Patent Application
20090172674
This invention relates, in general, to managing customer environments to provide support for business resiliency, and in particular, to managing the collection of information in an Information Technology environment to ensure the processing of requests for the information has minimal impact on the environment, but provides the most appropriate information to be used in managing the environment.
Today, customers attempt to manually manage and align their availability management with their information technology (IT) infrastructure. Changes in either business needs or the underlying infrastructure are often not captured in a timely manner and require considerable rework, leading to an inflexible environment.
Often high availability solutions and disaster recovery technologies are handled via a number of disparate point products that target specific scopes of failure, platforms or applications. Integrating these solutions into an end-to-end solution is a complex task left to the customer, with results being either proprietary and very specific, or unsuccessful.
Customers do not have the tools and infrastructure in place to customize their availability management infrastructure to respond to failures in a way that allows for a more graceful degradation of their environments. As a result, more drastic and costly actions may be taken (such as a site switch) when other options (such as disabling a set of applications or users) could have been offered, depending on business needs.
Coordination across availability management and other systems management disciplines is either nonexistent or accomplished via non-reusable, proprietary, custom technology.
There is little predictability as to whether the desired recovery objective will be achieved, prior to time of failure. There are only manual, labor intensive techniques to connect recovery actions with the business impact of failures and degradations.
Any change in the underlying application, technologies, business recovery objectives, resources or their interrelationships require a manual assessment of impact to the hand-crafted recovery scheme.
Based on the foregoing, a need exists for a capability to facilitate management of an IT environment. In one example, a need exists for a capability that manages the collection of information usable in managing the environment. As one example, a batch of requests (or queries) is executed within an allocated time interval. Then, depending on whether responses were received for the requests, processing associated with the batch of requests is adjusted in real-time to improve batch execution in a next iteration.
The shortcomings of the prior art are overcome and additional advantages are provided through the provision of a computer-implemented method to manage the collection of information in an Information Technology (IT) environment. The method includes, for instance, executing a batch of queries within an allocated time period; determining, in response to completion of the allocated time period, whether a response was not obtained for one or more queries of the batch of queries; and dynamically adjusting, in real-time, processing associated with the batch of queries, in response to the determining.
Computer program products and systems relating to one or more aspects of the present invention are also described and claimed herein.
Additional features and advantages are realized through the techniques of the present invention. Other embodiments and aspects of the invention are described in detail herein and are considered a part of the claimed invention.
One or more aspects of the present invention are particularly pointed out and distinctly claimed as examples in the claims at the conclusion of the specification. The foregoing and other objects, features, and advantages of the invention are apparent from the following detailed description taken in conjunction with the accompanying drawings in which:
In managing a customer's environment, such as its business environment, there is a set of requirements unaddressed by existing technology, which causes unpredictable down time, large impact failures and recoveries, and significant extra labor cost, with resulting loss of business revenue. These requirements include, for instance:
The above set of requirements is addressed, however, by a Business Resiliency (BR) Management System, of which one or more aspects of the present invention are included. The Business Resiliency Management System provides, for instance:
One goal of the BR system is to allow customers to align their supporting information technology systems with their business goals for handling failures of various scopes, and to offer a continuum of recovery services from finer grained process failures to broader scoped site outages. The BR system is built around the idea of identifying the components that constitute a business function, and identifying successive levels of recovery that lead to more complex constructs as the solution evolves. The various recovery options are connected by an overall BR management capability that is driven by policy controls.
Various characteristics of one embodiment of a BR system include:
A Business Resilience System is capable of being incorporated in and used by many types of environments. One example of a processing environment to incorporate and use aspects of a BR system, including one or more aspects of the present invention, is described with reference to
Processing environment 100 includes, for instance, a central processing unit (CPU) 102 coupled to memory 104 and executing an operating system 106. Examples of operating systems include AIX® and z/OS®, offered by International Business Machines Corporation; Linux; etc. AIX® and z/OS® are registered trademarks of International Business Machines Corporation, Armonk, N.Y., U.S.A. Other names used herein may be registered trademarks, trademarks or product names of International Business Machines Corporation or other companies.
The operating system manages execution of a Business Resilience Runtime Component 108 of a Business Resilience System, described herein, and one or more applications 110 of an application container 112.
As examples, processing environment 100 includes an IBM® System z™ processor or a pSeries® server offered by International Business Machines Corporation; a Linux server; or other servers, processors, etc. Processing environment 100 may include more, less and/or different components than described herein. (pSeries® is a registered trademark of International Business Machines Corporation, Armonk, N.Y., USA.)
Another example of a processing environment to incorporate and use aspects of a BR System, including one or more aspects of the present invention, is described with reference to
As shown, a processing environment 200 includes for instance, a central processing complex 202 coupled to an input/output (I/O) subsystem 204. Central processing complex 202 includes, for instance, a central processing unit 206, memory 208, an operating system 210, a database management system 212, a Business Resilience Runtime Component 214, an application container 216 including one or more applications 218, and an I/O facility 220.
I/O facility 220 couples central processing complex 202 to I/O subsystem 204 via, for example, a dynamic switch 230. Dynamic switch 230 is coupled to a control unit 232, which is further coupled to one or more I/O devices 234, such as one or more direct access storage devices (DASD).
Processing environments 100 and/or 200 may include, in other embodiments, more, less and/or different components.
In yet another embodiment, a central processing complex 300 (
For example, network service 302 of central processing complex 300 is coupled to a switch 308 of network subsystem 306. Switch 308 is coupled to a switch 310 via routers 312 and firewalls 314. Switch 310 is further coupled to a network service 316 of processing environment 304.
Processing environment 304 further includes, for instance, a central processing unit 320, a memory 322, an operating system 324, and an application container 326 including one or more applications 328. In other embodiments, it can include more, less and/or different components.
Moreover, CPC 300 further includes, in one embodiment, a central processing unit 330, a memory 332, an operating system 334, a database management system 336, a Business Resilience Runtime Component 338, an application container 340 including one or more applications 342, and an I/O facility 344. It also may include more, less and/or different components.
I/O facility 344 is coupled to a dynamic switch 346 of an I/O subsystem 347. Dynamic switch 346 is further coupled to a control unit 348, which is coupled to one or more I/O devices 350.
Although examples of various environments are provided herein, these are only examples. Many variations to the above environments are possible and are considered within the scope of the present invention.
In the above-described environments, a Business Resilience Runtime Component of a Business Resilience System is included. Further details associated with a Business Resilience Runtime Component and a Business Resilience System are described with reference to
In one example, a Business Resilience System 400 is a component that represents the management of recovery operations and configurations across an IT environment. Within that Business Resilience System, there is a Business Resilience Runtime Component (402) that represents the management functionality across multiple distinct Recovery Segments, and provides the service level automation and the support of creation of the recovery sequences. In addition, there are user interface (404), administration (406), installation (408) and configuration template (410) components within the Business Resilience System that enable the administrative operations that are to be performed. Each of these components is described in further detail below.
Business Resilience Runtime Component 402 includes a plurality of components of the BR System that are directly responsible for the collection of observations, creation of PSEs, policy acceptance, validation, error detection, and formulation of recovery sequences. As one example, Business Resilience Runtime Component 402 includes the following components:
In addition to the Business Resilience Runtime Component of the BR system, the BR system includes the following components, previously mentioned above.
The user interface, admin mailbox, install logic and/or template components can be part of the same computing unit executing BR Runtime or executed on one or more other distributed computing units.
To further understand the use of some of the above components and their interrelationships, the following example is offered. This example is only offered for clarification purposes and is not meant to be limiting in any way.
Referring to
As a result of these conditions leading up to runtime, the following subscriptions have already taken place:
These steps highlight one example of an error detection process:
In addition to the above, BR includes a set of design points that help in the understanding of the system. These design points include, for instance:
BR is targeted towards goal based policies—the customer configures his target availability goal, and BR determines the preparatory actions and recovery actions to achieve that goal (e.g., automatically).
Availability management of the IT infrastructure through goal based policy is introduced by this design. The BR system includes the ability to author and associate goal based availability policy with the resource Recovery Segments described herein. In addition, support is provided to decompose the goal policy into configuration settings, preparatory actions and runtime procedures in order to execute against the deployed availability goal. In one implementation of the BR system, the Recovery Time Objective (RTO—time to recover post outage) is a supported goal policy. Additional goal policies of data currency (e.g., Recovery Point Objective) and downtime maximums, as well as others, can also be implemented with the BR system. Recovery Segments provide the context for association of goal based availability policies, and are the scope for goal policy expression supported in the BR design. The BR system manages the RTO through an understanding of historical information, metrics, recovery time formulas (if available), and actions that affect the recovery time for IT resources.
RTO goals are specified by the customer at a Recovery Segment level and apportioned to the various component resources grouped within the RS. In one example, RTO goals are expressed as units of time intervals, such as seconds, minutes, and hours. Each RS can have one RTO goal per Pattern System Environment associated with the RS. Based on the metrics available from the IT resources, and based on observed history and/or data from the customer, the RTO goal associated with the RS is evaluated for achievability, taking into account which resources are able to be recovered in parallel.
Based on the RTO for the RS, a set of preparatory actions expressed as a workflow is generated. This preparatory workflow configures the environment or makes alterations in the current configuration, to achieve the RTO goal or to attempt to achieve the goal.
In terms of optimizing RTO, there are tradeoffs associated with the choices that are possible for preparatory and recovery actions. Optimization of recovery choice is performed by BR, and may include interaction at various levels of sophistication with IT resources. In some cases, BR may set specific configuration parameters that are surfaced by the IT resource to align with the stated RTO. In other cases, BR may request that an IT resource itself alter its management functions to achieve some portion of the overall RS RTO. In either case, BR aligns availability management of the IT resources contained in the RS with the stated RTO.
In this design, as one example, there is an approach to collecting the required or desired metrics data, both observed and key varying factors, system profile information that is slow or non-moving, as well as potential formulas that reflect a specific resource's use of the key factors in assessing and performing recovery and preparatory actions, historical data and system information. The information and raw metrics that BR uses to perform analysis and RTO projections are expressed as part of the IT resources, as resource properties. BR specific interpretations and results of statistical analysis of key factors correlated to recovery time are kept as BR Specific Management data (BRMD).
BR maintains specific information about the BR management of each resource pairing or relationship between resources. Information regarding the BR specific data for a resource pairing is kept by BR, including information such as ordering of operations across resources, impact assessment information, operation effect on availability state, constraint analysis of actions to be performed, effects of preparatory actions on resources, and requirements for resources to co-locate or anti-co-locate.
One feature of the BR function is the ability to identify the scope and impact of a failure. The BR design uses a Containment Region to identify the resources affected by an incident. The Containment Region is initially formed with a fairly tight restriction on the scope of impact, but is expanded on receiving errors related to the first incident. The impact and scope of the failure is evaluated by traversing the resource relationships, evaluating information on BR specific resource pairing information, and determining most current state of the resources impacted.
Various types of preparatory and recovery processes are formulated and in some cases, optionally initiated. Workflows used by BR are dynamically generated based on, for instance, customer requirements for RTO goal, based on actual scope of failure, and based on any configuration settings customers have set for the BR system.
A workflow includes one or more operations to be performed, such as Start CICS, etc. Each operation takes time to execute and this amount of time is learned based on execution of the workflows, based on historical data in the observation log or from customer specification of execution time for operations. The workflows formalize, in a machine readable, machine editable form, the operations to be performed.
In one example, the processes are generated into Business Process Execution Language (BPEL) compliant workflows with activities that are operations on IT resources or specified manual, human activities. For example, BRM automatically generates the workflows in BPEL. This automatic generation includes invoking routines to insert activities to build the workflow, or forming the activities and building the XML (Extensible Mark-Up Language). Since these workflows are BPEL standard compliant, they can be integrated with other BPEL defined workflows which may incorporate manual activities performed by the operations staff. These BR related workflows are categorized as follows, in one example:
Since the set of BR actions described above modify existing IT environments, visibility to the actions that are taken by BR prior to the actual execution is provided. To gain trust in the decisions and recommendations produced by BR, the BR System can run in ‘advisory mode’. As part of advisory mode, the possible actions that would be taken are constructed into a workflow, similar to what would be done to actually execute the processes. The workflows are then made visible through standard workflow authoring tooling for customers to inspect or modify. Examples of BPEL tooling include:
BR tooling spans the availability management lifecycle from definition of business objectives, IT resource selection, availability policy authoring and deployment, development and deployment of runtime monitors, etc. In one example, support for the following is captured in the tooling environment for the BR system:
The policy lifecycle for BR goal policies, such as RTO goals, includes, for example:
One of the points in determining operational state of a Recovery Segment is that this design allows for customers to configure a definition of specific ‘aggregated’ states, using properties of individual IT resources. A Recovery Segment is an availability management context, in one example, which may include a diverse set of IT resources.
The customer may provide the rules logic used within the Recovery Segment to consume the relevant IT resource properties and determine the overall state of the RS (available, degraded and unavailable, etc). The customer can develop and deploy these rules as part of the Recovery Segment availability policy. For example, if there is a database included in the Recovery Segment, along with the supporting operating system, storage, and network resources, a customer may configure one set of rules that requires that the database must have completed the recovery of in-flight work in order to consider the overall Recovery Segment available. As another example, customers may choose to configure a definition of availability based on transaction rate metrics for a database, so that if the rate falls below some value, the RS is considered unavailable or degraded, and evaluation of ‘failure’ impact will be triggered within the BR system. Using these configurations, customers can tailor both the definitions of availability, as well as the rapidity with which problems are detected, since any IT resource property can be used as input to the aggregation, not just the operational state of IT resources.
Failures occurring during sequences of operations executed within a BPEL compliant process workflow are intended to be handled through use of BPEL declared compensation actions, associated with the workflow activities that took a failure. The BR System creates associated “undo” workflows that are then submitted to compensate, and reset the environment to a stable state, based on where in the workflow the failure occurred.
The following set of customer values, as examples, are derived from the BR system functions described above, listed here with supporting technologies from the BR system:
Management of the IT environment is adaptively performed, as described herein and in a U.S. patent application “Adaptive Business Resiliency Computer System for Information Technology Environments,” (POU920070364US1), Bobak et al., co-filed herewith, which is hereby incorporated herein by reference in its entirety.
Many different sequences of activities can be undertaken in creating a BR environment. The following represents one possible sequence; however, many other sequences are possible. This sequence is provided merely to facilitate an understanding of a BR system and one or more aspects of the present invention. This sequence is not meant to be limiting in any way. In the following description, reference is made to various U.S. patent applications, which are co-filed herewith.
On receiving the BR and related product offerings, an installation process is undertaken. Subsequent to installation of the products, a BR administrator may define the configuration for BR manager instances with the aid of BRM configuration templates.
Having defined the BRM configuration a next step could be to define Recovery Segments as described in “Recovery Segments for Computer Business Applications,” (POU920070108US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Definition of a RS may use a representation of resources in a topology graph as described in “Use of Graphs in Managing Computing Environments,” (POU920070112US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
It is expected that customers will enable BR operation in “observation” mode for a period of time to gather information regarding key metrics and operation execution duration associated with resources in a RS.
At some point, sufficient observation data will have been gathered or a customer may have sufficient knowledge of the environment to be managed by BR. A series of activities may then be undertaken to prepare the RS for availability management by BR. As one example, the following steps may be performed iteratively.
A set of functionally equivalent resources may be defined as described in “Use of Redundancy Groups in Runtime Computer Management of Business Applications,” (POU920070113US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Specification of the availability state for individual resources, redundancy groups and Recovery Segments may be performed as described in “Use of Multi-Level State Assessment in Computer Business Environments,” (POU920070114US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Representations for the IT environment in which BR is to operate may be created from historical information captured during observation mode, as described in “Computer Pattern System Environment Supporting Business Resiliency,” (POU920070107US1), Bobak et al., which is hereby incorporated herein by reference in its entirety. These definitions provide the context for understanding how long it takes to perform operations which change the configuration—especially during recovery periods.
Information on relationships between resources may be specified based on recommended best practices—expressed in templates—or based on customer knowledge of their IT environment as described in “Conditional Computer Runtime Control of an Information Technology Environment Based on Pairing Constructs,” (POU920070110US1), Bobak et al., which is hereby incorporated herein by reference in its entirety. Pairing processing provides the mechanism for reflecting required or desired order of execution for operations, the impact of state change for one resource on another, the effect execution of an operation is expected to have on a resource state, desire to have one subsystem located on the same system as another and the effect an operation has on preparing the environment for availability management.
With preliminary definitions in place, a next activity of the BR administrator might be to define the goals for availability of the business application represented by a Recovery Segment as described in “Programmatic Validation in an Information Technology Environment,” (POU920070111US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Managing the IT environment to meet availability goals includes having the BR system prioritize internal operations. The mechanism utilized to achieve the prioritization is described in “Serialization in Computer Management,” (POU920070105US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Multiple operations are performed to prepare an IT environment to meet a business application's availability goal or to perform recovery when a failure occurs. The BR system creates workflows to achieve the required or desired ordering of operations, as described in “Dynamic Generation of Processes in Computing Environments,” (POU920070123US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
A next activity in achieving a BR environment might be execution of the ordered set of operations used to prepare the IT environment, as described in “Dynamic Selection of Actions in an Information Technology Environment,” (POU920070117US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Management by BR to achieve availability goals may be initiated, which may initiate or continue monitoring of resources to detect changes in their operational state, as described in “Real-Time Information Technology Environments,” (POU920070120US1), Bobak et al., which is hereby incorporated herein by reference in its entirety. Monitoring of resources may have already been initiated as a result of “observation” mode processing.
Changes in resource or redundancy group state may result in impacting the availability of a business application represented by a Recovery Segment. Analysis of the environment following an error is performed. The analysis allows sufficient time for related errors to be reported, insures gathering of resource state completes in a timely manner and insures sufficient time is provided for building and executing the recovery operations—all within the recovery time goal, as described in “Management Based on Computer Dynamically Adjusted Discrete Phases of Event Correlation,” (POU920070119US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
A mechanism is provided for determining if events impacting the availability of the IT environment are related, and if so, aggregating the failures to optimally scope the outage, as described in “Management of Computer Events in a Computer Environment,” (POU920070118US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Ideally, current resource state can be gathered after scoping of a failure. However, provisions are made to insure management to the availability goal is achievable in the presence of non-responsive components in the IT environment, as described herein, in accordance with one or more aspects of the present invention.
With the outage scoped and current resource state evaluated, the BR environment can formulate an optimized recovery set of operations to meet the availability goal, as described in “Defining a Computer Recovery Process that Matches the Scope of Outage,” (POU920070124US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Formulation of a recovery plan is to uphold customer specification regarding the impact recovery operations can have between different business applications, as described in “Managing Execution Within a Computing Environment,” (POU920070115US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Varying levels of recovery capability exist with resources used to support a business application. Some resources possess the ability to perform detailed recovery actions while others do not. For resources capable of performing recovery operations, the BR system provides for delegation of recovery if the resource is not shared by two or more business applications, as described in “Conditional Actions Based on Runtime Conditions of a Computer System Environment,” (POU920070116US1), Bobak et al., which is hereby incorporated herein by reference in its entirety.
Having evaluated the outage and formulated a set of recovery operations, the BR system resumes monitoring for subsequent changes to the IT environment.
In support of mainline BR system operation, there are a number of activities including, for instance:
In order to build a BR environment that meets recovery time objectives, IT configurations within a customer's location are to be characterized and knowledge about the duration of execution for recovery time operations within those configurations is to be gained. IT configurations and the durations for operation execution vary by time, constituent resources, quantity and quality of application invocations, as examples. Customer environments vary widely in configuration of IT resources in support of business applications. Understanding the customer environment and the duration of operations within those environments aids in insuring a Recovery Time Objective is achievable and in building workflows to alter the customer configuration of IT resources in advance of a failure and/or when a failure occurs.
A characterization of IT configurations within a customer location is built by having knowledge of the key recovery time characteristics for individual resources (i.e., the resources that are part of the IT configuration being managed; also referred to as managed resources). Utilizing the representation for a resource, a set of key recovery time objective (RTO) metrics are specified by the resource owner. During ongoing operations, the BR manager gathers values for these key RTO metrics and gathers timings for the operations that are used to alter the configuration. It is expected that customers will run the BR function in “observation” mode prior to having provided a BR policy for availability management or other management. While executing in “observation” mode, the BR manager periodically gathers RTO metrics and operation execution durations from resource representations. The key RTO metrics properties, associated values and operation execution times are recorded in an Observation log for later analysis through tooling. Key RTO metrics and operation execution timings continue to be gathered during active BR policy management in order to maintain currency and iteratively refine data used to characterize customer IT configurations and operation timings within those configurations.
Examples of RTO properties and value range information by resource type are provided in the below table. It will be apparent to those skilled in the art that additional, less, and/or different resource types, properties and/or value ranges may be provided.
A specific example of key RTO properties for a z/OS® image is depicted in
The z/OS® image has a set of RTO metrics associated therewith, as described above. Other resources may also have its own set of metrics. An example of this is depicted in
Further, in one example, the RTO properties from each of the resources that are part of the Recovery Segment for App A have been gathered by BR and formed into an “observation” for recording to the Observation log, as depicted at 850.
Resources have varying degrees of functionality to support RTO goal policy. Such capacity is evaluated by BR, and expressed in resource property RTOGoalCapability in the BRMD entry for the resource. Two options for BR to receive information operation execution timings are: use of historical data or use of explicitly customer configured data. If BR relies on historical data to make recovery time projections, then before a statistically meaningful set of data is collected, this resource is not capable of supporting goal policy. A mix of resources can appear in a given RS—some have a set of observations that allow classification of the operation execution times, and others are explicitly configured by the customer.
Calculation of projected recovery time can be accomplished in two ways, depending on customer choice: use of historical observations or use of customers input timings. The following is an example of values for the RTOGoalCapability metadata that is found in the BRMD entry for the resource that indicates this choice:
If the customer is in observation mode, then historical information is captured, regardless of whether the customer has indicated use of explicitly input timings or use of historical information.
The administrator can alter, on a resource basis, which set of timings BR is to use. The default is to use historical observations. In particular, a change source of resource timing logic is provided that alters the source that BR uses to retrieve resource timings. The two options for retrieving timings are from observed histories or explicitly from admin defined times for operation execution. The default uses information from the observed histories, gathered from periodic polls. If the customer defines times explicitly, the customer can direct BR to use those times for a given resource. If activated, observation mode continues and captures information, as well as running averages, and standard deviations. The impact to this logic is to alter the source of information for policy validation and formulation of recovery plan.
With respect to the historical observations, there may be a statistically meaningful set of observations to verify. The sample size should be large enough so that a time range for each operation execution can be calculated, with a sufficient confidence interval. The acceptable number of observations to qualify as statistically meaningful, and the desired confidence interval are customer configurable using BR UI, but provided as defaults in the BRMD entry for the resource. The default confidence interval is 95%, in one example.
There are metrics from a resource that are employed by BR to enable and perform goal management. These include, for instance:
There is also a set of information about the resource that is employed—this information is provided as defaults in the BRMD entry for the resource, but provided to the BR team in the form of best practices information/defaults by the domain owners:
In addition to the resources defined herein as part of the IT configuration that is managed, there are other resources, referred to herein as assessed resources. Assessed resources are present primarily to provide observation data for PSE formation, and to understand impact(s) on managed resources. They do not have a decomposed RTO associated with them nor are they acted on for availability by BR. Assessed resources have the following characteristics, as examples:
Similarly, there are likely scenarios where a resource exists in a customer environment that already has an alternative availability management solution, and does not require BR for its availability. However, since other resources that are managed by BR may be dependent on them, they are observed and assessed in order to collect observation data and understand their impacts on managed resources. Additionally, there may be resources that do not have alternative management solutions, but the customer simply does not want them managed by BR, but other managed resources are dependent upon them. They too are classified as assessed resources.
These assessed resources share many of the same characteristics of managed resources, such as, for example:
Finally, there are a few restrictions that BR imposes upon assessed resources, in this embodiment:
To facilitate the building of the customer's IT configuration, observations regarding the customer's environment are gathered and stored in an observation log. In particular, the observation log is used to store observations gathered during runtime in customer environments, where each observation is a collection of various data points. They are created for each of the Recovery Segments that are in “observation” mode. These observations are used for numerous runtime and administrative purposes in the BR environment. As examples the observations are used:
BR gathers observations during runtime when “observation mode” is enabled at the Recovery Segment level. There are two means for enabling observation mode, as examples:
The administrator may also disable observation mode for a Recovery Segment, which stops it from polling for data and creating subsequent observation records for insertion in the log. However, the accumulated observation log is not deleted. In one example, an RS remains in observation mode throughout its lifecycle. The UI displays the implications of disabling observation mode.
In BR, the observations that are collected by BR during runtime can be grouped into two categories, as examples:
A periodic poll observation is a point-in-time snapshot of the constituent resources in a Recovery Segment. Observation data points are collected for those resources in the Recovery Segment(s) which have associated BR management data for any of the following reasons, as examples:
The full value of these observations is derived for an RS when they include data that has been gathered for its constituent resources, plus the resources that those are dependent upon. In one embodiment, the administrator is not forced to include all dependent resources when defining a Recovery Segment, and even if that were the case, there is nothing that prevents them from deleting various dependent resources. When defining a Recovery Segment, the BR UI provides an option that allows the customer to display the dependency graph for those resources already in the Recovery Segment. This displays the topology from the seed node(s) in the Recovery Segment down to and including the dependent leaf nodes. The purpose of this capability is to give the customer the opportunity to display the dependent nodes and recommend that they be included in the Recovery Segment.
Preparatory and recovery workflows are built by the BR manager to achieve the customer requested RTO policy based on resource operations timings. During active policy monitoring by the BR manager, measurements of achieved time for operations are recorded in observations to the log and used to maintain the running statistical data on operation execution times. Observations written to the log may vary in the contained resource RTO metrics and operation execution timings.
Observations are also collected from any of the BPEL workflows created by BR in the customer's environment. There is a standard template that each BR BPEL workflow uses. As part of that template, observation data is captured at the start of, during, and at the completion of each workflow. Specifically, in one example, one observation is created at the end of the workflow with data accumulated from completion of each activity. This information is used to gather timings for workflow execution for use in creating subsequent workflows at time of failure.
In accordance with an aspect of the present invention, management of an IT environment, such as an IT environment that supports Business Resiliency, is facilitated by the controlled gathering of information used to manage the environment.
Today, Business Resilience technologies typically rely solely on the reliability of incoming event processing, and do very little (if any) collection of state during normal operations for querying or ascertaining the state of resources. In addition, if such queries are performed, there are no components to allow for the collection of this information in a manner that minimizes overhead and ensures ability to meet required goals, such as a Recovery Time Objective. The drawbacks include potentially stale (and inaccurate) information to be used in making recovery decisions; queries that are initiated that cause delay beyond what the recovery time tolerance will allow for a business application; and failure to initiate assessment of state during normal operations to determine an expected level of performance for resources during various times of the day.
A technique of distributing queries asynchronously is provided herein, in which the underlying services being invoked support synchronous behavior (i.e., once a query or request is submitted, the process does nothing until a response is returned). (In this embodiment, the services do not support asynchronous behavior (i.e., after submission of a query, the process continues performing other actions and does not wait for a response in order to proceed). However, in another embodiment, both synchronous and asynchronous behaviors are supported.) The queries are distributed via a distributor. The queries support a tolerance for wait time which is dependent on the context of the invocation, and the distributor further parallelizes the queries across the set of input resources. Both the technique used to invoke the distributor, along with the processing within the distributor, are covered by this process. Various characteristics associated with this process include, for instance, a parallelized asynchronous distributor; wait tolerance in context of invocation; minimization of performance impact; adjustment of microintervals; handling of a responses missing timeout window; response handling; and local optimizations, each of which is described below.
In one embodiment, the Asynchronous Distributor of this process parallelizes queries to underlying services that are synchronous in nature. The problem with a large set of synchronous services that need to be invoked is the performance impact of waiting for each successive response, when processing potentially multiple thousands of requests. The wait time for responses could exceed what can be tolerated by many applications, including applications that manage the infrastructure, such as for business resiliency. The asynchronous distributor described herein accepts a batch of requests from any client invocation, and in this case, the business resiliency management components, and parallelizes each request of the batch to run on a separate thread. The request to the service itself is synchronous, as that is what the service supports. However, across the batch of requests, the threads parallelize the queries. In a complex environment, the expectation is to have multiple asynchronous distributors, placed in a locally optimized way. The services that each distributor has local optimization capability for are kept by the invoker, so queries can be directed to the appropriate asynchronous distributor.
One example of a high level view of an asynchronous distributor 900 is depicted in
BR has a design point for scale that is targeted to the large, complex environment. During recovery processing, BR expects that potentially a large set of resources are impacted. In large z/OS® Sysplex environments, it is not unusual to expect anywhere from 500,000 to 1,000,000 resources distributed across 25 WebSphere containers (assuming each WAS container supports 25,000 instances). In cases where a large number of these resources are to be queried within a short period of time, it is impractical to try to accomplish this in a synchronous manner. In fact, synchronous query during a recovery process that is time sensitive will be an issue even for just a single query.
Further details relating to BRAD and
One of the areas of difficulty in synchronous behavior of the underlying services is the expected or allowed time for wait. By definition, synchronous services return when they have completed, either successfully or unsuccessfully. They are not constrained by a time period, but instead, are considered as time independent. When there are critical time dependent invocations of large sets of these services, the wait time cannot be predicted or guaranteed to complete within a given window. The asynchronous distributor described herein accepts from its caller a time sensitive context that allows each thread mentioned in item (1) to be allocated a timeout. In this manner, the caller's tolerance for wait time is applied to the query. The distributor explicitly sets a timer around the query invocation and if the timer expires prior to query completion for that individual query, the response is returned as null. The timeout is on an individual basis, so if the batch contains 100 requests, and 97 complete in the allocated time, and 3 of them do not complete, the other 97 still contain response information. The timeout is not fixed by the distributor, but can vary by the invoker on each call to the distributor.
Using this technique, the underlying synchronous services operate within a time sensitive bound, and processing of the client using the distributor (in this case, business resiliency) can explicitly have control over whether ‘sufficient’ data has been received for the allowed time, or whether additional queries have to be initiated to the same or alternate interfaces to determine the information. Business resilience uses this distributor in multiple contexts, both for collecting observations during normal operations, as well as during recovery time error assessment processing.
One of the goals of the asynchronous distributor is the minimization of performance impact to the invoker and to the overall system during the query processing. In some cases, the invoker requires responses as soon as available, but large spikes in performance can be caused by submitting a significant number of parallel queries in a small interval of time. As a result, the invoker methodology used by business resilience varies depending on the context of the invocation.
Normal observation invocation:
Error assessment invocation:
In cases where business resilience uses the distributors for normal observations, there is an explicit wait between batches to ensure the spread of requests over the complete interval. However, in some cases, processing for the complete set of requests may not complete in time, and in some cases, processing for the complete set of requests may complete in less time than the complete interval. The business resilience component that invokes the distributors for normal observations measures the time to invoke the complete set of distributor requests and adjusts the wait interval between batches (the microinterval), along with the time that each query is allocated to complete accordingly based on responses. In this manner, the batching of queries and the thread timeout used by the distributors are dynamically adjusted continuously to optimize for minimum performance overhead to the system.
In some cases, processing the complete set of requests to the distributors may not complete in the allocated time. The business resilience component invoking the distributor tracks the missed responses and calculates a running average of request to response percentage. Notification is then sent to the administrator so that intervals can be adjusted if necessary, or problems with repeatedly slow responding resources can be investigated.
The requests to the distributors are, for instance, asynchronous, and each distributor sends responses back to the invoker for each batch that is to be processed. The responses back to the invoker from the distributors are parallelized, and may occur out of order since the invocation is asynchronous. Tokens are used as part of the request and response to correlate the response back and ensure that any time sensitive query responses are associated with the correct observation or discarded, if more recent information has been received on a more current error assessment. For performance reasons related to locking of database records, the business resilience components centralize update of runtime management information by the invoker, after the asynchronous distributors have all responded, for a given query or state assessment.
The design for asynchronous collection of information works optimally when the asynchronous distributors themselves can be placed in a manner that is optimized with the services that they will be asked to invoke. Although that is not a strict requirement of the design, it is a further optimization that is incorporated by the business resilience design. Lists of which services are hosted by a given application server, on a given OS, can be programmatically collected and maintained, and the invocation logic apportions requests to the asynchronous distributors based on those services that are most local to each distributor. The services that are part of a batch request that comes to a distributor from a business resilience invocation for normal observation or state assessment is based on programmatic inspection of the list that identifies services and where the services are hosted. The distributors are deployed into the same environment for the services to which they will initiate query requests. In this way, network communication costs and full marshalling/demarshalling costs between the distributor and each of the parallelized queries can be avoided. Because of the nature of the services being invoked in the case of business resilience, the service request will not fail if the distributor that invokes the service is not local, but rather executes in a non-optimized manner. As a result, the business resilience design places the distributors in an optimized manner in the environment so that invoked services are localized as much as possible.
The BRAD EJB may be implemented as an EJB 3.0 stateless session bean so that multiple BRAD clients can simultaneously access it and invoke methods on it. The Java beans that comprise the EJB itself execute within an EJB container, such as the IBM® WebSphere Application Server (WAS). The BRAD clients may optionally reside and execute within an EJB container, but are not specifically required to do so. The EJB is to have both a local and a remote interface so that the clients can invoke operations on it either remotely if they are in the different EJB containers (or on different servers) or locally if they are in the same EJB container (or on the same physical server).
The BRAD EJB is deployed and resides within each of the WAS containers in the BR environment that hosts resources. This allows BR to take advantage of the local optimization available within the same container when invoking operations on resources. In one implementation, the BRAD is written in Java, which allows Java-to-Java communications via Remote Method Invocation (RMI). The parameters passed on the requests/responses are internal to the BRAD mechanism, and are therefore, optimized, for example, by eliminating the need to marshal, unmarshal, and parse XML files.
Logically, there are three components to the BRAD, as depicted in
1. BRAD Client 902
2. Distributor 908
3. Query Thread 914
The list of resource representations running in each WAS container are maintained at the Recovery Segment level in a RS.BRAD_List. That list is initialized when the Recovery Segment is defined and associated with a particular BRM via the BR UI. This list only pertains to the resources in the Recovery Segment though, not all the resources in the environment. For every constituent resource in the Recovery Segment there is a corresponding entry in the list that indicates the WAS server and hosting container, which may be derived from a JMX interface (described, e.g., in Java™ and JMX: Building Manageable Systems, Heather Kreger, Ward Harold, Leigh Williamson, Addison-Wesley Professional, Jan. 9, 2003 (ISBN-10: 0672324083; ISBN-13: 978-0672324086); and Java Management Extensions, J. Seven Perry, O'Reilly Media, Inc., 1st edition, Jun. 15, 2002 (ISBN-10: 0596002459; ISBN-13: 978-0596002459), each of which is hereby incorporated herein in its entirety) to each of the WAS containers hosting resource representations. It is also the responsibility of the Recovery Segment to keep the RS.BRAD_List current in case a resource is moved, or one is encountered in the RS but does not contain a corresponding WAS hosting entry in the list. If a new resource is encountered in the RS that is hosted in a container without a deployed BRAD EJB in it, a notification is sent to the administrator's mailbox indicating that one is to be deployed.
As previously mentioned, each distributor EJB uses a fixed-set thread pool for requests. The exact number of threads used is governed by the number of resources that have to be queried and the amount of time the requester allows for it to complete its allotted work. During recovery time, the list of resources is provided from the BRM for Containment Region(s) and the tolerance for delay is calculated based upon the timing framework and will likely be very short, which forces a larger sized pool of threads. During observation mode, the Recovery Segment provides a list of resources to the distributors and evenly staggers the number of calls to each distributor based upon the number of resources in the RS.BRAD_List, and the amount of time per periodic interval (which will likely be considerably longer since the default is 15 minutes). Additionally, a pacing technique in the BRAD client logic continuously adjusts the response tolerance and number of resources to batch per request based upon the elapsed time of previous requests. The number of resources batched per request starts with a default of 20, but adjusts slightly higher or lower as necessary or desired with each periodic interval. If the RS eventually determines that sufficient data is not being collected per observation to be useful, a notification is sent to the administrator's mailbox indicating that the polling interval may be too small and should be increased.
With reference to
A BradEjbBean class 1000 implements the JAVA SessionBean interface class. It may be implemented as a singleton class to ensure there is only a single instance for each resource hosting container. Since it implements the SessionBean interface class, in one example, it implements the following operations and attributes, which are specific to session beans, not to BR.
The following operations and attributes are specific to the BR implementation (also shown in the figure):
A BradDistributor class 1002 encapsulates the functions used to communicate with the BRAD clients and the resources that are to be queried. Except for accepting requests from BRAD clients, the method invocations on the distributor are driven by the BradEjb class. Thus, all the knowledge entailed with which resources to query when can be encapsulated in the BradEjb, and the distributor only handles the various communications with the clients and the resources. In one example, it is implemented as a singleton class so that there is only a single instance for each BRAD EJB, and should be instantiated during the init( ) method of the BradEjb class. One example of the BradDistributor operations and attributes is described below:
A BradClient class 1004 is used to communicate to the BRAD EJB, and can utilize either the remote or local homes of the EJB. One embodiment of the BradClient operations and attributes is described below:
During runtime periodic poll, observations are gathered from the environment by each Recovery Segment that is in observation mode. The data that is to be collected for a periodic poll are, for instance:
The data is used to populate the observation record and to maintain cached values in RS and BRMD for usage during recovery failures. The Recovery Segment interactions with the BR Asynchronous Distributor in the environment are described in more detail in the example below.
The interaction pattern for the BR Manager with the distributors is a bit different than that of the Recovery Segment. The BR Manager interacts with the BRAD distributors during a time of failure when expediency is very important.
In one implementation, another BRAD interaction is with the z/OS® Workload Manager (WLM) offered by International Business Machines Corporation. An interface to WLM is provided through the z/OS® OperatingSystem resource representation. WLM provides various RTO and performance metrics for CPU, memory, and I/O consumption and delays for a given set of address spaces. In general, BR interfaces to WLM in the following manner:
On the next observation for that z/OS® OperatingSystem resource, that Token is passed back to WLM (via, for instance, the stopApplicationRecoveryMonitor(Token) operation). That accomplishes two things: first, it stops the sampling for those subsystems; and second, it retrieves the RTO and performance metrics gathered by WLM for those address spaces during the periodic poll interval.
The BRAD client logic at the Recovery Segment parses the data based on the mapping information provided by the z/OS® OperatingSystem resource, and updates the corresponding entries in the observation record (for those corresponding subsystem resources) prior to the insertion of the record into the observation log. That data is also saved in the BRMD entry for the Recovery Segment.
A periodic poll observation is a point-in-time snapshot of the constituent resources in a Recovery Segment. Observation data points are collected, in one embodiment, for those resources in the Recovery Segment(s) which have associated BR management data for any of the following reasons:
The full value of these observations is derived for a RS when they contain data that has been gathered for its constituent resources, plus the resources that those are dependent upon. Currently, the administrator is not forced to include all dependent resources when defining a Recovery Segment, and even if that were the case, there is nothing that prevents them from deleting various dependent resources. Currently, BR employs a number of best-practices techniques to assist the customer in configuring BR for runtime monitoring and management. A similar technique is implemented in the BR UI to assist the customer for observation mode. Customers are able to define Recovery Segments through the usage of definition templates (which IBM® recommends as a best-practice), or alternatively they may configure Recovery Segments manually. In either case, when defining a Recovery Segment, the BR UI provides an option that allows the customer to display the dependency graph for those resources already in the Recovery Segment. This displays the topology from the seed node(s) in the Recovery Segment down to and including all the dependent leaf nodes. The purpose of this capability is to give the customer the opportunity to display the dependent nodes and recommend that they be included in the Recovery Segment. As an example, a dependency graph 1300 for a Recovery Segment 1302 might look like the graph depicted in
If the customer selects to accept the recommended proposal, the RS is then expanded to include all the resources, not just the small set originally selected, as shown. Obviously, if the customer chooses not to invoke the UI option (i.e., “Display Dependency Graph”), or chooses not to accept the recommended proposal, the RS is not expanded. However, the administrator is alerted of the subsequent implications of not doing so, and advised against it.
In one example, there may be a set of resources within a Recovery Segment for which the customer added specifically for the purposes of monitoring and management by BR (via a goal policy), and another distinct set that the customer does not desire BR to manage, but still is to be “observed” by BR in order to collect the necessary or desired information to properly manage the managed set of resources for availability. As a result, a new class of resource has been defined to describe these observed resources termed, assessed resources.
While in observation mode, the Recovery Segment is responsible for periodically polling the relevant BR Asynchronous Distributors (BRADs) in the environment for the resources in the Recovery Segment. The RS provides the list of resources for each BRAD to query and an observation token so that the multiple observation records can be correlated together from the BR UI. Each BRAD then invokes the necessary operations on the resources in the list provided by the Recovery Segment, aggregates the responses into a single observation record, and returns it to the Recovery Segment for insertion into the observation log and for updating the metadata associated with the resources via the BRMD tables.
An overview of the periodic poll process is described with reference to
The interval for performing periodic observations is based on, for instance, the RS resource property (PERIODIC_POLL_INVERVAL) that is configured via the BR UI. The staggering and pacing of the observation data is governed at the BRAD (via a thread pool to achieve parallelism based on the number of resources to query). The idea again is to not overwhelm or in any way degrade the system with the collection and storing of these observations. Note that since the observation timestamp is calculated by the RS at the end of each interval based on the current value of the PERIODIC_POLL_INVERVAL, it automatically accommodates any UI adjustments to it by the administrator (i.e., increasing or decreasing the interval). Finally, if the interval is altered to an unrealistically small (e.g., 1 minute) or large value (e.g., 999 minutes), the administrator is warned of the implications and advised against such an alteration.
Adjustments are made to four factors, as an example, to optimize processing, meet periodic poll interval requirements and minimize overhead of the periodic poll process. These include, for instance, adjustment to initiation cycle of periodic poll; invocation of requests for resources not responding; alteration of number of query threads; and alteration of number of requests per batch, batch size and pacing time for batches.
These processes work in conjunction with each process having an impact on others which is synergistically managed by the overall BRAD process.
The specified periodic poll interval is used as a staring point in determining the timing of batches. The number of requests per batch and the number of resources represented in the RS determines the number of batches. Based on the number of batches and the periodic poll interval, a microinterval for each batch is calculated, as described below. Actual time for the process may be longer or shorter than expected due to delays in request/response processing, delays in responses from resources and processing time for the technique. At the end of the periodic poll cycle, the actual time to complete the cycle is calculated. A ratio of the actual time to the desired periodic poll interval is calculated and used to scale the target periodic poll interval, also described below. Note that the target periodic poll interval is used as the reference point. The scaled periodic poll interval used in the technique is adjusted based on runtime characteristics of the system where the reality of the processing is empirically measured and compensated for by scaling the periodic poll interval for the next cycle.
Responses from BRAD processing include information from resources and an indication if a response from the resource was received before the microinterval timeout. On the next invocation of the BRAD from the periodic poll initiation process, those resources for which a response was not received are processed for threadpool execution first, as described below. Resources which responded in the last periodic poll cycle are processed and made available for threadpool execution after the resources which did not respond. This gives the non-responsive resources from the previous cycle priority and the full microinterval to complete as they have access to the threadpool first.
The number of query threads in the threadpool is initialized based on BR distributed calculations. The number of resources responding and the proportion of the interval used to receive the responses are used to adjust the threadpool size. If all resources have responded and no more than, for instance, 70% of the available interval time has been utilized, the number of threads in the threadpool is decreased, as described below. The threadpool is contracted at a rate of, for instance, 10% of the threads. This is a slow contraction process which requires multiple iterations to shrink the number of threads by half. If all resources have not responded, the number of threadpool threads may be increased. The increase is, for instance, half the percentage of the difference between the number of requests in the batch and the number of resources not responding, as described below. If the percent of response not received is less than, for instance, 10%, the threadpool is set to its maximum size which is equal to the number of requests in the batch. This is a relative rapid increase in the number of threads in order to quickly meet the needs of periodic poll processing. It is paced by the previous number of threads, the number of requests completing and the number of requests not providing a response. Therefore, it adjusts to start increasing rapidly when needed and slow as the target of completing all requests is approached. The limit to the increase in threads in the threadpool is the total number of requests in a batch. At that point, each request is initiated as soon as the cycle begins and has the full microinterval to complete.
Adjustments to the number of requests in a batch is paced to work synergistically with the adjustments to the threadpool number of threads. No change is made to the number of requests in a batch for, for instance, the first 10 iterations of the periodic poll cycle. If at the end of 10 cycles there are resources which are not providing a response, the threadpool adjustments to the number of threads will have practically reached stabilization at the number of threads per batch. The number of requests per batch is adjusted by ⅓, in one example, as described below. This lengthens the microinterval by ⅓ allowing for a larger proportion of resource requests to complete. Adjusting the number of requests per batch also drives the threadpool number of threads processed. If the increased number of requests are not completing, the threadpool has an increased number of threads. Consumption of the additional 33% of requests per periodic poll cycle requires the threadpool number of threads routine approximately 4 cycles to reach maximum threads per threadpool. Therefore, the number of requests per batch is adjusted at most every fourth poll cycle, in this embodiment.
If all batches of requests respond before expiration of the allotted portion of the poll interval, the number of batches may be increased with a corresponding decrease in the number of requests per batch. The total of the not used by requests, i.e., the time difference between that in which the response arrived and the allotted portion of the poll interval, is maintained as responses to requests are received. The minimum response time is also maintained as responses to requests are received, as described below. If the total of the unused time is greater than the smallest response time, the number of batches is increased by one with a corresponding decrease in the number of requests per batch, also described below.
While one embodiment of the above technique is described in the following logic, extensions or alterations are achievable. For example:
In the following logic, references are made to services for thread pool management, such as create a thread pool (new); and remove a thread pool (shutdownNow). Descriptions for these services can be found in, for instance, ISBN 0131482025: Core Java™ 2, Volume I—Fundamentals (7th Edition) (Core Series), and ISBN 0131118263: Core Java™ 2, Volume II—Advanced Features (7th Edition) (core Series), each of which is hereby incorporated herein by reference in its entirety.
Deployment of BRAD in Associate RS with BRM
Establishing the BR environment includes interaction with the BR administrator for deployment of BRAD functionality. One step in creation of the BR environment insures the existence of an association of resources being managed and a BRAD instance. Optimally, the BRAD associated with a resource instance is within the same hosting environment enabling low overhead for requests presented from the BRAD to the resource representation for data.
One embodiment of the logic to deploy a BRAD is described with reference to
Referring to
A determination is made regarding the hosting environment for the resource representation, STEP 1502. In one implementation, this information may be provided through a UI interaction with the customer. The UI interaction may enable a group of resources to be identified as associated with one hosting container, in one example. Another implementation may invoke a JMX interface to determine the hosting environment.
Further, a determination is made as to whether a BRAD is deployed to the hosting container, STEP 1504. In one implementation, this is accomplished through a UI interaction with the customer where BR administrator assurance of BRAD deployment is obtained. In another implementation, software interfaces may be utilized to determine if BRAD functionality has been deployed.
If a BRAD has not been deployed, INQUIRY 1506, a request to the BR administrator is made via the UI to cause a BRAD to be made operational in the hosting environment, INQUIRY 1508. If the BR administrator does not deploy a BRAD, a UI interaction may request specification of an alternate BRAD to be used to gather data on the resource, STEP 1510.
When a BRAD has been established for the resource (Y from INQUIRY 1508, or Y from INQUIRY 1506, or from STEP 1510), an entry in the RS.BRAD_List is made that includes the identification of the resource and the associated BRAD identification, STEP 1512. Processing then continues at STEP 1500.
The list of resources associated with a BRAD built during RS deployment is updated during ongoing systems operation. In one implementation, the RS.BRAD_List may be updated when:
Update to the RS.BRAD_List may be performed as a complete refresh or as a selective alteration. A similar process to establishing the RS.BRAD_List as RS deployment time is followed for a single resource update or for a complete refresh. During runtime, updates can be performed without involvement of the BR administrator, if programming interfaces exist for determining the hosting container for a resource representation and for deploying a BRAD in a hosting container. If programming interfaces do not exist for those functions, notification is provided to the BR administrator through the mailbox.
BRAD initialization is initiated by, for instance, WAS when the BRAD EJB is started in the WAS container. Two threadpools are allocated, a fixed-size threadpool for periodic observations and a cached threadpool for state queries. The fixed-size threadpool is better suited for queries that are not as time sensitive as the state queries, since it dispatches fewer threads simultaneously, whereas the cached threadpool dispatches the number of threads required immediately.
One embodiment of the logic to initialize a BRAD is described with reference to
Referring to
Two threadpools are created. One of a fixed size for the periodic observations; the other is a cached thread pool for the query states. The threadpools serve as the BRAD dispatcher. The fixed size threadpool has active the number of threads allocated for it. When more tasks are submitted than threads available, they are queued up by the threadpool manager of WAS. As threads free up, tasks are read off the queue provided by the WAS threadpool manager.
The cached threadpool dispatches everything submitted to it through WAS services using free threads if available or allocating new threads. There are no tasks that are not immediately dispatched to the cached thread pool, in this implementation.
In one example, a count of threads for the fixed pool is calculated from the total number of resources divided by 1000 with the result incremented by one, STEP 1610. This number may be adjusted during ongoing operation of the BRAD. If the fixed pool number of threads is less than, for instance, 10, INQUIRY 1620, the count is set to be 10, in this example, STEP 1622. Otherwise, or if set to 10, a fixed thread pool is created through invocation of WAS services with the calculated thread count, STEP 1624.
Additionally, a cached thread pool is created through invocation of WAS services with an unbounded number of threads in order to immediately begin processing of all tasks associated with a query state request, STEP 1626.
Having initialized the BRAD, no further processing is performed until a request for BRAD processing is received as, for example, from periodic poll or state query processing.
At each polling interval, BR sends a query to the set of resources managed for a given RS to collect, for instance, state, RTO metrics, operation execution timings, properties associated with 1st level state aggregation rules, and properties associated with triggers for pairing rules. Roundtrip times and clock variations are also recorded. A part of the information collected is recorded into the Observation Log and part is used to update the BRMD and BRRD information. The observation collection is phased across resources over the polling interval, parallelized and made asynchronous to achieve minimal performance impact.
This logic is initiated when the UI user sets the observation mode resource property for a selected RS, or when policy has been activated. The periodic poll process is operated continuously during BR runtime. Adjustments are made to the number of requests in a batch and the wait time for a batch(s) of requests based on observed response completions, timeouts and time to respond. All requests to the BRADs are performed asynchronously.
One embodiment of the logic to initiate periodic poll observation is described with reference to
Referring to
If a BRAD is associated with the resource and the associated BRAD is not temporary, INQUIRY 1712, processing proceeds to determine if the BRAD is already in the ByBRADResList, STEP 1721 (
The ByBRADResList is built to include a list of resources associated with each BRAD providing service to the RS. If the BRAD associated with the resource has already been placed in the ByBRADResList, INQUIRY 1721 (
Returning to INQUIRY 1721, if the BRAD has not been placed in the list, a new row is created in the ByBRADResList, STEP 1724, the BRAD associated with the resource is stored in the new row, STEP 1726, and the next resource column index in the new row is set to 1, STEP 1728. Having set or retrieved the next column index for the row, the resource associated with the BRAD is stored in the ByBRADResList, STEP 1730, the next column index for the row is incremented by one, STEP 1731, and processing continues at STEP 1710 (
When all resources associated with the RS have been processed, STEP 1710, initialization for the poll cycle is executed. The RS.ObservationToken is set to reflect the current poll cycle, STEP 1732 (
Data to be used in making dynamic adjustments to the number of requests per batch is initialized for the poll cycle. The number of requests to BRAD(s) is set to zero, as are the number of resources providing responses to the poll cycle, PerObs_Res_Response, and the accumulated wait time across all of the batch MicroInterval(s), STEP 1738. Recording of the minimum time any one request to a BRAD required, RespMin, is used to determine if the number of batches should be decreased. It is initialized to the length of the polling cycle, so that as responses arrive, the minimum of the current response and RespMin can be kept as the running minimum response time, STEP 1740.
The ByBRADResList is prepared for poll cycle processing. Each row of the ByBRADResList, STEP 1742, has the last resource column set based on the next resource column index created when adding resources to a row, STEP 1744. The next resource to be processed for a BRAD is initialized to the first column index (e.g., 1), STEP 1746.
An indication, AllSent, is set to false reflecting that all requests for BRAD(s) have not been sent, STEP 1748 (
For the poll cycle, a cycle is executed (e.g., STEPs 1750-1769, described below), in which a batch of requests is sent to a BRAD and the MicroInterval for the batch is exhausted before the next request is sent. Requests are sent sequentially to each BRAD in the ByBRADResList, so long as there exist resource(s) for that BRAD which have not been processed in this poll cycle. A determination is made if all requests have been sent for this poll cycle, INQUIRY 1750. If not, the indication that all requests have been sent is set true, STEP 1751. This indication is reset if any requests are sent when processing through the ByBRADResList. An index for moving through the ByBRADResList is initialized to 1, STEP 1752. A loop through the ByBRADResList is performed with checking to determine when all entries in the ByBRADResList have been processed, INQUIRY 1753. When the last row in the ByBRADResList has been processed, the next iteration through the ByBRADResList may be performed INQUIRY 1750. Otherwise, a comparison of the next resource column to the last resource column for the row determines if there are resources associated with this BRAD for which a request has not been made this poll cycle, INQUIRY 1754. If remaining resources do not exist for this BRAD, the next BRAD is processed, STEP 1755. Otherwise, the indication of all processing having been performed is set to false, STEP 1756.
A request for a BRAD is created (e.g., STEPs 1757-1768). The number of requests made to BRAD(s) for this poll cycle is incremented by one, STEP 1757 (
Identification of the BRAD to which the request for resource data is to be sent is set in the request message, STEP 1763. The resource identification for each resource in the batch is moved from the ByBRADResList to the request message, RequestMsg, STEP 1764. The token for this poll cycle (established at STEP 1732) is set in the request message, ObservationToken, for correlation when a response is received, which is also used to determine if delayed responses are to be discarded, STEP 1765 (
When all batches for the poll cycle have been processed, INQUIRY 1750, statistics for the periodic poll process are generated. The ending time is set to the current TOD, STEP 1770 (
If the elapsed time for this polling cycle is less than the desired interval, INQUIRY 1782, a delay equal to the difference is introduced, STEP 1784. Otherwise, or when the delay completes, the next periodic poll cycle begins, INQUIRY 1702 (
When periodic polling is to stop, INQUIRY 1702, the time for the current cycle is saved in the RS, STEP 1704 (
The BRAD logic for when a periodic poll observation request type is sent to the BRAD EJB is initiated by a BRAD client request from one of the WAS containers in the BR environment. Input includes, for instance, a request type (RequestType), a token representing the observation instance so the client can correlate responses with the request, a list of the resources and the data to be retrieved from the resource (ResList), and a time within which a response is required (MaxResponseTime). There are tasks that are submitted to the threadpool that get dispatched as threads become available. A response message is created based on the input request message, populated and then returned.
The assumption on the observation request is that the caller has provided all the necessary information in the ResList to do multiple queries to the same resource if required (e.g., to query an RTO metric, but also to query for state on that resource).
One embodiment of the logic to process BRAD Requests is described with reference to
Referring to
For a periodic poll process with no active state queries in process, INQUIRY 1808, a data structure, RequestHash, is created, STEP 1812. Each thread initiated to make a resource request updates an entry in the RequestHash array corresponding to the resource for which data was retrieved. The RequestHash structure includes, for instance, the ObservationToken from the request message, a StateArray which has a row for each resource from which data is to be requested, and a count of responses which have been placed in the structure, which is initialized to 0. An interval timer is set to the maximum time allotted to this batch of requests based on the MaxResponseTime contained in the request message, STEP 1814. At expiration of the interval timer, the BRAD_Response routine is to be given control.
For each resource in the request message ResList, STEP 1816, if the resource did not provide a response to the last poll cycle, INQUIRY 1818, a thread pool work element is submitted, STEP 1820, identifying the resource from the request message, an index into the StateArray for the output of the thread, which is the same index as the index into the request message ResList, and an anchor for the RequestHash shared data structure. This initial pass through the input prioritizes requests to resources which did not respond in the last periodic poll cycle. Processing continues at STEP 1816.
Returning to INQUIRY 1818, if the resource did provide a response, processing continue at STEP 1816.
After the first pass through the input, a second pass through each entry in the input message ResList is made, STEP 1822 (
Returning to INQUIRY 1824, if the resource did not respond, processing continues at STEP 1822.
When all resources in the input request message have been formed into thread pool work elements, processing completes until the next BRAD request message is received.
The BRAD state query logic executes when a state query request type is sent to the BRAD EJB. It is initiated by a BRAD client request from one of the WAS containers in the BR environment. Tasks are submitted to the threadpool that are dispatched immediately.
One embodiment of the BRAD state query logic is described with reference to
Referring to
The BRAD query thread logic describes the process taken by the JAVA threads that actually do the synchronous queries of the resources. They are initiated by the threadpool dispatcher when work elements have been submitted to the threadpool. In one implementation, the resource and property/operation are provided on request and there could be multiples of each which would allow the usage of getMultipleResourceProperty. In the implementation described herein, single requests for individual property/value(s) are processed.
One embodiment of the BRAD query thread logic is described with reference to
Referring to
If the response count matches the number of resource requests which were to be processed, INQUIRY 2010, the interval timer for the allotted time for all requests to complete is cancelled, STEP 2012, and the BRAD_Response routine is invoked, STEP 2014. Subsequently, or if all data has not been retrieved, INQUIRY 2010, the thread is freed for processing the next threadpool work element, STEP 2016.
The BRAD response logic is given control when all responses have been populated in the RequestHash or on expiration of the timer for the maximum time allotted for this BRAD request batch.
One embodiment of the BRAD Response logic is described with reference to
Referring to
If a response to a state query is being processed, INQUIRY 2110, the number of active state queries is decreased by one, STEP 2112. If the active state query count indicates there are no requests in process, INQUIRY 2116 (
Returning to INQUIRY 2100 (
Processing to adjust the thread pool executes (STEPs 2123-2156). If all requests completed before expiration of the allotted interval and less than, for instance, 70% of the interval was used, INQUIRY 2124 (
If processing of the current batch required more than 70% of the allotted time, INQUIRY 2124 (
If the percent of resource not providing a response is greater than, for instance, 10%, INQUIRY 2140, a thread count increase is set to half of the percent not responding, STEP 2142 (
Returning to INQUIRY 2140 (
BRAD client completion processes the response message from BRAD(s) which have retrieved resource data. The response message includes, for instance, the observation token identifying the periodic poll cycle or state query request, property value for resource(s) and the request TOD for when the request was originated to the BRAD.
One embodiment of the BRAD client completion processing is described with reference to
Referring to
Otherwise, statistics are generated for BRAD processing (e.g., STEPS 2204-2212). The number of responses received for this iteration of BRAD requests is incremented by one, STEP 2204. The current TOD is saved, STEP 2206, and used to calculate the response time for this request by subtracting the request TOD returned in the response message, STEP 2208. The maximum response time for a request for resource data is saved in the RS, RS.Level_T2_interval_max, STEP 2210. The number of responses received for the current execution of the BRAD process is updated with the count of resources responding in the current response message, STEP 2212.
Data is maintained for subsequent use in adjusting the number of requests per batch in STEPS 2214-2220 (
If all requests for this cycle of BRAD processing have been submitted (RS.ObsDone—YES) and all response messages have been received, INQUIRY 2222, processing continues. Otherwise, processing ends.
When all responses to a BRAD processing cycle have been received, statistics are created. The percent of responses received across all requests is calculated from the sum of resources responding divided by the number of resources in the RS, STEP 2224 (
Processing to determine which resources responded in the current BRAD cycle is performed (e.g., STEPS 2228-2234). For each resource in the current batch, i.e., each StateArray entry in the response message, STEP 2228, the value in the response is checked for being null. If the resource provided a response, INQUIRY 2230, the corresponding BRAD_list entry is marked as having a response, STEP 2232. Otherwise, the BRAD_list entry is marked as not having a response, STEP 2234. In either case, processing continues at STEP 2228.
Adjustment of the number of requests in a batch, and therefore, the number of batches and time allotted for each batch is performed (e.g., STEPS 2238-2256). If, for instance, 10 or fewer periodic poll cycles have executed for the RS, INQUIRY 2238 (
Returning to INQUIRY 2240, if all requests are completing in the current BRAD cycle, it may be possible to decrease the number of requests in a batch in order to spread out processing and achieve a more consistent, less disruptive periodic poll process. The accumulated wait time for the batch is compared to the smallest response for a request within the batch, INQUIRY 2250 (
Described in detail herein is a capability for dynamically managing the processing associated with executing requests to obtain information usable in managing an IT environment.
One or more aspects of the present invention can be included in an article of manufacture (e.g., one or more computer program products) having, for instance, computer usable media. The media has therein, for instance, computer readable program code means or logic (e.g., instructions, code, commands, etc.) to provide and facilitate the capabilities of the present invention. The article of manufacture can be included as a part of a computer system or sold separately.
One example of an article of manufacture or a computer program product incorporating one or more aspects of the present invention is described with reference to
A sequence of program instructions or a logical assembly of one or more interrelated modules defined by one or more computer readable program code means or logic direct the performance of one or more aspects of the present invention.
Advantageously, a capability is provided for managing in real-time the gathering of information to be used in managing aspects of an Information Technology (IT) environment. Processing associated with the execution of a batch of requests within an allotted time frame is adjusted in real-time, in response to a determination of whether responses were received for the requests. Advantageously, the time period for executing requests can be adjusted, as well as the number of requests in a batch and the priority of the requests in the batch. Advantageously, at least a portion of the requests are executed concurrently.
Although various embodiments are described above, these are only examples. For example, the processing environments described herein are only examples of environments that may incorporate and use one or more aspects of the present invention. Environments may include other types of processing units or servers or the components in each processing environment may be different than described herein. Each processing environment may include additional, less and/or different components than described herein. Further, the types of central processing units and/or operating systems or other types of components may be different than described herein. Again, these are only provided as examples.
Moreover, an environment may include an emulator (e.g., software or other emulation mechanisms), in which a particular architecture or subset thereof is emulated. In such an environment, one or more emulation functions of the emulator can implement one or more aspects of the present invention, even though a computer executing the emulator may have a different architecture than the capabilities being emulated. As one example, in emulation mode, the specific instruction or operation being emulated is decoded, and an appropriate emulation function is built to implement the individual instruction or operation.
In an emulation environment, a host computer includes, for instance, a memory to store instructions and data; an instruction fetch unit to obtain instructions from memory and to optionally, provide local buffering for the obtained instruction; an instruction decode unit to receive the instruction fetched and to determine the type of instructions that have been fetched; and an instruction execution unit to execute the instructions. Execution may include loading data into a register for memory; storing data back to memory from a register; or performing some type of arithmetic or logical operation, as determined by the decode unit. In one example, each unit is implemented in software. For instance, the operations being performed by the units are implemented as one or more subroutines within emulator software.
Further, a data processing system suitable for storing and/or executing program code is usable that includes at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements include, for instance, local memory employed during actual execution of the program code, bulk storage, and cache memory which provide temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
Input/Output or I/O devices (including, but not limited to, keyboards, displays, pointing devices, DASD, tape, CDs, DVDs, thumb drives and other memory media, etc.) can be coupled to the system either directly or through intervening I/O controllers. Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modems, and Ethernet cards are just a few of the available types of network adapters.
Further, although the environments described herein are related to the management of availability of a customer's environment, one or more aspects of the present invention may be used to manage aspects other than or in addition to availability. Further, one or more aspects of the present invention can be used in environments other than a business resiliency environment.
Yet further, many examples are provided herein, and these examples may be revised without departing from the spirit of the present invention. For example, in one embodiment, the description is described in terms of availability and recovery; however, other goals and/or objectives may be specified in lieu of or in addition thereto. Additionally, the resources may be other than IT resources. Further, there may be references to particular products offered by International Business Machines Corporation or other companies. These again are only offered as examples, and other products may also be used. Additionally, although tables and databases are described herein, any suitable data structure may be used. There are many other variations that can be included in the description described herein and all of these variations are considered a part of the claimed invention.
Further, for completeness in describing one example of an environment in which one or more aspects of the present invention may be utilized, certain components and/or information is described that is not needed for one or more aspects of the present invention. These are not meant to limit the aspects of the present invention in any way.
One or more aspects of the present invention can be provided, offered, deployed, managed, serviced, etc. by a service provider who offers management of customer environments. For instance, the service provider can create, maintain, support, etc. computer code and/or a computer infrastructure that performs one or more aspects of the present invention for one or more customers. In return, the service provider can receive payment from the customer under a subscription and/or fee agreement, as examples. Additionally or alternatively, the service provider can receive payment from the sale of advertising content to one or more third parties.
In one aspect of the present invention, an application can be deployed for performing one or more aspects of the present invention. As one example, the deploying of an application comprises providing computer infrastructure operable to perform one or more aspects of the present invention.
As a further aspect of the present invention, a computing infrastructure can be deployed comprising integrating computer readable code into a computing system, in which the code in combination with the computing system is capable of performing one or more aspects of the present invention.
As yet a further aspect of the present invention, a process for integrating computing infrastructure, comprising integrating computer readable code into a computer system may be provided. The computer system comprises a computer usable medium, in which the computer usable medium comprises one or more aspects of the present invention. The code in combination with the computer system is capable of performing one or more aspects of the present invention.
The capabilities of one or more aspects of the present invention can be implemented in software, firmware, hardware, or some combination thereof. At least one program storage device readable by a machine embodying at least one program of instructions executable by the machine to perform the capabilities of the present invention can be provided.
The flow diagrams depicted herein are just examples. There may be many variations to these diagrams or the steps (or operations) described therein without departing from the spirit of the invention. For instance, the steps may be performed in a differing order, or steps may be added, deleted, or modified. All of these variations are considered a part of the claimed invention.
Although embodiments have been depicted and described in detail herein, it will be apparent to those skilled in the relevant art that various modifications, additions, substitutions and the like can be made without departing from the spirit of the invention and these are therefore considered to be within the scope of the invention as defined in the following claims.
