Patent Application
20160103914
This application is one of several U.S. Nonprovisional patent applications filed contemporaneously. The related applications are ROW-LEVEL SECURITY INTEGRATION OF ANALYTICAL DATA STORE WITH CLOUD ARCHITECTURE (Atty. Docket No. SALE 1096-1/1451US), LOW LATENCY ARCHITECTURE WITH DIRECTORY SERVICE FOR INTEGRATION OF TRANSACTIONAL DATA SYSTEM WITH ANALYTICAL DATA STRUCTURES (Atty. Docket No. SALE 1098-1/1453US), INTEGRATION USER FOR ANALYTICAL ACCESS TO READ ONLY DATA STORES GENERATED FROM TRANSACTIONAL SYSTEMS (Atty. Docket No. SALE 1099-1/1454US), VISUAL DATA ANALYSIS WITH ANIMATED INFORMATION MORPHING REPLAY (Atty. Docket No. SALE 1100-1/1455US), DECLARATIVE SPECIFICATION OF VISUALIZATION QUERIES DISPLAY FORMATS AND BINDINGS (Atty. Docket No. SALE 1101-1/1456US) and DASHBOARD BUILDER WITH LIVE DATA UPDATING WITHOUT EXITING AN EDIT MODE (Atty. Docket No. SALE 1103-1/1458US). The related applications are hereby incorporated by reference for all purposes.
The subject matter discussed in the background section should not be assumed to be prior art merely as a result of its mention in the background section. Similarly, a problem mentioned in the background section or associated with the subject matter of the background section should not be assumed to have been previously recognized in the prior art. The subject matter in the background section merely represents different approaches, which in and of themselves may also correspond to implementations of the claimed technology.
The advent of powerful servers, large-scale data storage and other information infrastructure has spurred the development of advance data warehousing and data analytics applications. Structured query language (SQL) engines, on-line analytical processing (OLAP) databases and inexpensive large disk arrays have for instance been harnessed to capture and analyze vast streams of data. The analysis of that data can reveal valuable trends and patterns not evident from more limited or smaller-scale analysis.
In the case of transactional data management, the task of inspecting, cleaning, transforming and modeling data with the goal of discovering useful information is particularly challenging due to the complex relationships between different fields of the transaction data. Consequently, performance of conventional analytical tools with large transaction data sets has been inefficient. That is also in part because the time between requesting a particular permutation of data and that permutation's availability for review is directly impacted by the extensive compute resources required to process standard data structures. This heavy back-end processing is time-consuming and particularly burdensome to the server and network infrastructure.
The problem is worsened when an event occurs that renders the processing interrupted or stopped. In such an event, latency is incurred while waiting for the processing to re-initiate so that the appropriate action takes place. This latency is unacceptable for analytics applications that deliver real-time or near real-time reports. Accordingly, systems and methods that can alleviate the strain on the overall infrastructure are desired.
An opportunity arises to provide business users full ad hoc access for querying large-scale database management systems and rapidly building analytic applications by using efficient queueing protocols for faster creation and processing of massively compressed datasets. Improved customer experience and engagement, higher customer satisfaction and retention, and greater sales may result.
In the drawings, like reference characters generally refer to like parts throughout the different views. Also, the drawings are not necessarily to scale, with an emphasis instead generally being placed upon illustrating the principles of the technology disclosed. In the following description, various implementations of the technology disclosed are described with reference to the following drawings, in which:
The technology disclosed relates to integration between large-scale transactional systems and temporary analytic data stores suitable for use by one or more analysts. In other implementations, the technology disclosed relates to integration between large-scale transactional systems, non-structured data stores (e.g., log files), analytical systems (corporate data warehouse, department data marts), and personal data sources (spreadsheets, csv files).
Exploration of data without updating the underlying data presents a different use case than processing transactions. A data analyst may select, organize, aggregate and visualize millions or even hundreds of millions of transactional or log records without updating any of the records. So-called EdgeMart™ analytic data store technology, developed by EdgeSpring®, has been demonstrated to manipulate 123 million Federal Aviation Administration (FAA) records, on a laptop running a browser, with sub-one second response time for processing a query, including grouping, aggregation and result visualization. Storing the underlying records in a read only purpose designed analytic data structure makes these results possible using modest hardware. Producing, managing and operating analytic data stores at scale remains challenging.
Analytic data structures, also referred to as “edgemart(s),” are compressed data forms produced from transactional databases, which represent specific form functions of transactional database objects. Sometimes analytic data structures are produced by merging data from multiple database systems or platforms. For instance, prospect and opportunity closing data may come from a Salesforce.com® system and order fulfillment data from a SAP® system. An analytic data structure may combine sales and fulfillment data for particular opportunities, merging data from systems that run on different database platforms, in separate applications from different vendors, applying divergent security models. Dozens of analysts may work on subsets of an overall analytic data structure, both for periodic and ad hoc investigations. Their work is likely to be directed to a specific time period, such as last month, last quarter or the last 30 days. Different requirements of analysts can be accommodated using technology disclosed herein.
There are many aspects to addressing the challenge of scaling an analytic system architecture that draws from large scale transactional systems. First, the resources needed can be reduced by using a purposed designed low-latency messaging protocol between transactional system components and analytic data store components. Second, divergent security models of multiple transactional systems can be addressed by a predicate-based row-level security scheme capable of translating various security settings for use in an analytic data store. Security can be arranged in a manner that facilitates building individual shards of an analytical data store for users who either want or have access limited to a particular segment of the overall data.
Third, operation of an analytic data store can be facilitated by a separate accounting of analytic resource usage. The technology disclosed keeps the analytic resource usage accounting separate by associating a so-called integration user for analytic services with a standard transactional user. Transactional user credentials and processing of authentication and authorization can be leveraged to invoke the associated integration user. This associated user has different rights and different accounting rules that the transactional user.
Fourth, migration of query processing from servers to clients can mitigate high peak loads followed by idle periods observed when delivering extremely fast data exploration and visualization. The technology disclosed further includes a strategy for migration, during a particular investigation session, of query processing from server based to client based.
Low latency communication between a transactional system and analytic data store resources can be accomplished through a low latency key-value store with purpose-designed queues and status reporting channels. Posting by the transactional system to input queues and complementary posting by analytic system workers to output queues is described. On-demand production and splitting of analytic data stores requires significant elapsed processing time, so a separate process status reporting channel is described to which workers can periodically post their progress, thereby avoiding progress inquiries and interruptions of processing to generate report status. This arrangement produces low latency and reduced overhead for interactions between the transactional system and the analytic data store system.
A directory service associated queuing and transactional system to worker inter-process communications enables restarting of worker processes running on analytic system servers that fail. Workers running on separate servers and even in separate server racks are redundantly assigned affinities to certain queues and clients. When one of the redundant workers fails and restarts, the directory service provides information so that status and task information can be obtained by the restarted worker from the redundant sister workers. This keeps the workers from recreating edgemart(s) that were created while the worker was off-line, according to one implementation.
A predicate-based row level security system is used when workers build or split an analytical data store. According to one implementation, predicate-based means that security requirements of source transactional systems can be used as predicates to a rule base that generates one or more security tokens, which are associated with each row as attributes of a dimension. Similarly, when an analytic data store is to be split, build job, user and session attributes can be used to generate complementary security tokens that are compared to security tokens of selected rows. Efficient indexing of a security tokens dimension makes it efficient to qualify row retrieval based on security criteria.
Building analytical data stores from transactional data systems that have divergent security models is facilitated by predicate-based rules that translate transactional security models and attributes into security tokens, according to one implementation. For instance, Saleforce.com® allows a tenant to select among about seven different security models. Selecting any one of these models could make it difficult or impossible to express security requirements expressed according to a different model. Selecting one of the Salesforce.com® models could complicate expressing security requirements implemented under an SAP® security model. Predicate-based rules facilitate extracting data objects consistent with needs of analytical data structure users. A single analytical data store can be built for sharing among multiple users and for providing security consistent with underlying security models and analytical data access rights of users. Security tokens can be assigned to rows based on criteria such as “CEOs can access all transactional records for the last five years,” which might not be implemented or expressed in the underlying transactional systems. It is expected that analysts will have access to records for analytical purposes that they might not be allowed to or might find cumbersome to access through the underlying transactional systems.
Splitting an analytical data store refers to creating a so-called shard, which is a second analytical data store created by selecting a proper subset of data objects or rows in a first analytical data store. This can be regularly scheduled, alongside refreshing of an analytical data store with updated data from the transactional data system. Or, it can happen on demand or on an ad hoc basis. The technology disclosed can be applied to create shards from larger analytical data stores. In one implementation, creating a subset of an edgemart for simultaneous storage and subsequent deployment along with the original edgemart is referred to as “physical splitting.” In some implementations, physically splitting of edgemart(s) is performed over-night or through batch processing. In such implementations, the resulting shards are stored in a cache and are made available on-demand in response to user queries. In another implementation, providing a subset of data stored in an edgemart in response to a query without maintaining a separate subset edgemart is referred to as “logical splitting.” In the logical splitting implementation, deployment of the subset of the edgemart's data is qualified based on authentication and authorization of a user who initiated the query.
Creating shards can be beneficial for regularly scheduled creation of analytical data stores, especially when production involves creation of multiple data stores with overlapping data. It has been observed that creation of user-requested, specific data stores can be brittle in the sense of easily breaking People leave and join analytical groups. Jobs are created and then forgotten. Underlying data changes. When dozens or hundreds of analytical data stores derive from a single shared set of data, process brittleness can be reduced by hierarchical creation of analytical data stores. A predicate-based row level security rule set facilitates hierarchical data store assembly.
An automated, hierarchical process of creating even two hierarchical levels of analytical data stores can benefit from predicate-based row level security rules. At a first hierarchical level, security tokens can be created and associated at a row level with data objects. The security tokens can encode security attributes that facilitate creation of the second or subsequent hierarchical levels of analytical data stores, given the flexibility afforded by predicate-based rules. A three level creation system can have additional benefits, related to structuring of patterns of analytical data store creation. The relationship among analytical data store children created from a single mother analytical data store can be more clearly revealed by multiple generations of relationships that correspond to three or more hierarchical levels.
After creation of analytical stores, use of a so-called integration user can control access rights and be used for accounting. By its nature, a temporary analytical data store involves much more limited rights to modify or update data than typical in a transactional data system. A typical user may have read/search rights to at least one analytical data store. Even if the user has write/update writes to the transactional data system(s) from which the analytical data stores are created, the user may only have read/search rights. The user may further have recreate-on-demand rights, but the read only nature of the analytical data store makes it unnecessary for the user to enjoy the write/update rights that the user has with the corresponding transactional data system. Or, the user's analytical data store rights may be restricted to a first company subdivision, even if the user occasionally contributes to results in a second company subdivision. In some implementations, the integration user can be given rights under a predicate-based set of security rules, but this is not necessary.
The transactional user also can facilitate accounting for analytical data store usage. Use of analytical data stores for high performance data exploration typically involves a fraction of the user base size that generates transactions. As mentioned above, their data exploration generates much higher peak loads than individual transactions. These conditions are likely to lead to different licensing conditions for analytical data store system users than for transactional system users.
Again, the so-called integration user keeps the analytic resource usage accounting separate by associating an integration user for analytic services with a standard transactional user. Transactional user credentials and processing of authentication and authorization can be leveraged to invoke the associated integration user. Then, the associated user's rights and accounting rules can be applied to meet analytic security and accounting needs with minimal burdens on the pre-existing transactional system.
Aggressive exploration can involve multiple, successive queries and visualizations. This creates difficulty scaling the resources needed to deliver fast responses. It is particularly complicated by regular rebuilding of analytic data stores, whether daily or on demand. Migrating queries using the technology described involves migrating indexed fields, known as dimensions, and quantity fields, known as measures, in the background during a query session. A session that starts in server query processing mode may switch to client query processing as enough data fields have been copied from the server to the client. When the client determines that it has enough data fields to process an incoming query, it can locally process the new query without passing it to the server. Since both the server and client are working from copies of the same read only analytic data structure, a user receives the same results from either client or the server.
These features individually and collectively contribute to integration of an analytic data store system with one or more legacy transactional systems.
The described subject matter is implemented by a computer-implemented system, such as a software-based system, a database system, a multi-tenant environment, or the like. Moreover, the described subject matter can be implemented in connection with two or more separate and distinct computer-implemented systems that cooperate and communicate with one another. One or more implementations can be implemented in numerous ways, including as a process, an apparatus, a system, a device, a method, a computer readable medium such as a computer readable storage medium containing computer readable instructions or computer program code, or as a computer program product comprising a computer usable medium having a computer readable program code embodied.
Examples of systems, apparatus, and methods according to the disclosed implementations are described in a “transaction data” context. The examples of transaction data are being provided solely to add context and aid in the understanding of the disclosed implementations. In other instances, other data forms and types related to other industries like entertainment, animation, docketing, education, agriculture, sports and mining, medical services, etc. may be used. Other applications are possible, such that the following examples should not be taken as definitive or limiting either in scope, context, or setting. It will thus be apparent to one skilled in the art that implementations may be practiced in or outside the “transaction data” context.
As used herein, a given signal, event or value is “dependent on” a predecessor signal, event or value of the predecessor signal, event or value influenced by the given signal, event or value. If there is an intervening processing element, action or time period, the given signal, event or value can still be “dependent on” the predecessor signal, event or value. If the intervening processing element or action combines more than one signal, event or value, the signal output of the processing element or action is considered “dependent on” each of the signal, event or value inputs. If the given signal, event or value is the same as the predecessor signal, event or value, this is merely a degenerate case in which the given signal, event or value is still considered to be “dependent on” the predecessor signal, event or value. “Responsiveness” of a given signal, event or value upon another signal, event or value is defined similarly.
In analytics environment 100 a runtime framework with event bus 125 manages the flow of requests and responses between an explorer engine 102, a query engine 122 and a live dashboard engine 108. Data acquired (extracted) from large data repositories is used to create “raw” edgemart(s) 142—read-only data structures for analytics, which can be augmented, transformed, flattened, etc. before being published as customer-visible edgemart(s) for business entities. A query engine 122 uses optimized data structures and algorithms to operate on these highly-compressed edgemart(s) 142, delivering exploration views of this data. Accordingly, an opportunity arises to analyze large data sets quickly and effectively.
Visualization queries are implemented using a declarative language to encode query steps, widgets and bindings to capture and display query results in the formats selected by a user. An explorer engine 102 displays real-time query results. When activated by an analyst developer, explorer engine 102 runs EQL queries against the data and includes the data in lenses. A lens describes a single data visualization: a query plus chart options to render the query. The EQL language is a real-time query language that uses data flow as a means of aligning results. It enables ad hoc analysis of data stored in Edgemart(s). A user can select filters to change query parameters and can choose different display options, such as a bar chart, pie chart or scatter plot—triggering a real-time change to the display panel—based on a live data query using the updated filter options. An EQL script consists of a sequence of statements that are made up of keywords (such as filter, group, and order), identifiers, literals, or special characters. EQL is declarative: you describe what you want to get from your query. Then, the query engine will decide how to efficiently serve it.
A runtime framework with an event bus 125 handles communication between a user application 158, a query engine 122 and an explorer engine 102, which generates lenses that can be viewed via a display engine 118. A disclosed live dashboard engine 108 designs dashboards, displaying multiple lenses from the explorer engine 102 as real-time data query results. That is, an analyst can arrange display panels for multiple sets of query results from the explorer engine 102 on a single dashboard. When a change to a global filter affects any display panel on the dashboard, the remaining display panels on the dashboard get updated to reflect the change. Accurate live query results are produced and displayed across all display panels on the dashboard.
Explorer engine 102 provides an interface for users to choose filtering, grouping and visual organization options; and displays results of a live query requested by a user of the application 158 running on a user computing device 148. The query engine 122 executes queries on read only pre-packaged data sets—the edgemart data structures 142. The explorer engine 102 produces the visualization lens using the filter controls specified by the user and the query results served by the query engine 122.
Explorer engine 102, query engine 122 and live dashboard engine 108 can be of varying types including a workstation, server, computing cluster, blade server, server farm, or any other data processing system or computing device. In some implementations, explorer engine 102 can be communicably coupled to a user computing device 148 via different network connections, such as the Internet. In some implementations, query engine 122 can be communicably coupled to a user computing device 148 via different network connections, such as a direct network link. In some implementations, live dashboard engine 108 can be communicably coupled to user computing device 148 via different network connections, such as the Internet or a direct network link.
Runtime framework with event bus 125 provides real time panel display updates to the live dashboard engine 108, in response to query results served by the query engine 122 in response to requests entered by users of application 158. The runtime framework with event bus 125 sets up the connections between the different steps of the workflow.
Display engine 118 receives a request from the event bus 125, and responds with a first chart or graph to be displayed on the live dashboard engine 108. Segments of a first chart or graph are filter controls that trigger generation of a second query upon selection by a user. Subsequent query requests trigger controls that allow filtering, regrouping, and selection of a second chart or graph of a different visual organization than the first chart or graph.
Display engine 118 includes tweening engine 128 and tweening stepper 138 that work together to generate pixel-level instructions—intermediate frames between two images that give the appearance that the first image evolves smoothly into the second image. The drawings between the start and destination frames help to create the illusion of motion that gets displayed on the live dashboard engine 108 when a user updates data choices.
Runtime framework with event bus 125 can be of varying types including a workstation, server, computing cluster, blade server, server farm, or any other data processing system or computing device; and can be any network or combination of networks of devices that communicate with one another. For example, runtime framework with event bus 125 can be implemented using one or any combination of a LAN (local area network), WAN (wide area network), telephone network (Public Switched Telephone Network (PSTN), Session Initiation Protocol (SIP), 3G, 4G LTE), wireless network, point-to-point network, star network, token ring network, hub network, WiMAX, WiFi, peer-to-peer connections like Bluetooth, Near Field Communication (NFC), Z-Wave, ZigBee, or other appropriate configuration of data networks, including the Internet. In other implementations, other networks can be used such as an intranet, an extranet, a virtual private network (VPN), a non-TCP/IP based network, any LAN or WAN or the like.
Edgemart engine 152 uses an extract, load, transform (ELT) process to manipulate data served by backend system servers to populate the edgemart data structures 142. Edgemart data structures 142 can be implemented using a general-purpose distributed memory caching system. In some implementations, data structures can store information from one or more tenants into tables of a common database image to form an on-demand database service (ODDS), which can be implemented in many ways, such as a multi-tenant database system (MTDS). A database image can include one or more database objects. In other implementations, the databases can be relational database management systems (RDBMSs), object oriented database management systems (OODBMSs), distributed file systems (DFS), no-schema database, or any other data storing systems or computing devices.
In some implementations, user computing device 148 can be a personal computer, a laptop computer, tablet computer, smartphone or other mobile computing device, personal digital assistant (PDA), digital image capture devices, and the like. Application 158 can take one of a number of forms, including user interfaces, dashboard interfaces, engagement consoles, and other interfaces, such as mobile interfaces, tablet interfaces, summary interfaces, or wearable interfaces. In some implementations, it can be hosted on a web-based or cloud-based privacy management application running on a computing device such as a personal computer, laptop computer, mobile device, and/or any other hand-held computing device. It can also be hosted on a non-social local application running in an on premise environment. In one implementation, application 158 can be accessed from a browser running on a computing device. The browser can be Chrome, Internet Explorer, Firefox, Safari, and the like. In other implementations, application 158 can run as an engagement console on a computer desktop application.
In other implementations, environment 100 may not have the same elements or components as those listed above and/or may have other/different elements or components instead of, or in addition to, those listed above, such as a web server and a template database. The different elements or components can be combined into single software modules and multiple software modules can run on the same hardware.
Network(s) 225 is any network or combination of networks of devices that communicate with one another. For example, network(s) 225 can be any one or any combination of a LAN (local area network), WAN (wide area network), telephone network (Public Switched Telephone Network (PSTN), Session Initiation Protocol (SIP), 3G, 4G LTE), wireless network, point-to-point network, star network, token ring network, hub network, WiMAX, WiFi, peer-to-peer connections like Bluetooth, Near Field Communication (NFC), Z-Wave, ZigBee, or other appropriate configuration of data networks, including the Internet. In other implementations, other networks can be used such as an intranet, an extranet, a virtual private network (VPN), a non-TCP/IP based network, any LAN or WAN or the like.
In some implementations, the various engines illustrated in
In some implementations, a transaction data management system 232 can store structured, semi-structured, unstructured information from one or more tenants into tables of a common database image to form an on-demand database service (ODDS), which can be implemented in many ways, such as a multi-tenant database system (MTDS). A database image can include one or more database objects. In other implementations, the transaction data management system 232 can be a relational database management system (RDBMSs), an object oriented database management systems (OODBMSs), a distributed file systems (DFS), a no-schema database, or any other data storing system or computing device.
Client(s) 255 can communicate with various components of the integration environment 200 using TCP/IP and, at a higher network level, use other common Internet protocols to communicate, such as HTTP, FTP, AFS, WAP, etc. As an example, where HTTP is used, client(s) 255 can employ an HTTP client commonly referred to as a “browser” for sending and receiving HTTP messages from an application server included in the pod engines 222. Such application server can be implemented as the sole network interface between pod engines 222 and superpod engines 204, but other techniques can be used as well or instead. In some implementations, the interface between pod engines 222 and superpod engines 204 includes load sharing functionality 202, such as round-robin HTTP request distributors to balance loads and distribute incoming HTTP requests evenly over a plurality of servers in the integration environment.
In one aspect, the environment shown in
Queuing engine 208 defines a dispatching policy for the integration environment 200 to facilitate interactions between a transactional database system and an analytical database system. The dispatching policy controls assignment of requests to an appropriate resource in the integration environment 200. In one implementation of the dispatching policy, a multiplicity of messaging queues is defined for the integration environment, including a “named key-value task start queue” and a “named key-value task complete queue.” The “named key-value task start queue” dispatches user requests for information. The “named key-value task complete queue” dispatches information that reports completion of the user requests. In other implementations, when either the processing time exceeds the maximum response time or the size of the data set exceeds the data threshold, a progress report can be sent to the user. The progress reports refers to information transmitted to advise an entity of an event, status, or condition of one or more requests the entity initiated.
Application of the multiplicity of messaging queues solves the technical problem of queue blockage in the integration environment 200. Contention is created when multiple worker threads use a single queue to perform their tasks. Contention in multi-threaded applications of queues can slow down processing in the integration environment 200 up to three orders, thus resulting in high latency. The condition is worsened when there are multiple writers adding to a queue and readers consuming. As a result, every time a request is written or added to a particular queue, there is contention between multiple worker threads since a reader concurrently attempts to read or remove from the same queue. In some implementations, integration environment 200 uses a pool of worker threads for reading or writing requests from or to clients in the network(s) 225. Worker threads are hosted on resources referred to as “workers.” Once request is read into the “named key-value task start queue,” it is dispatched for execution in the workers. The resulting data generated after the request is executed by the workers is referred is stored as edgemart(s) 142. In some implementations, the edgemart(s) 142 are portioned into multiple smaller edgemart(s) called shards 216. In one implementation, edgemart(s) 142 are partitioned based on specified dimensions such as a range or a hash.
Various types of on-demand transactional data management systems can be integrated with analytic data stores to provide data analysts ad hoc access to query the transaction data management systems. This can facilitate rapid building of analytic applications that use numerical values, metrics and measurements to drive business intelligence from transactional data stored in the transaction data management systems and support organizational decision making Transaction data refers data objects that support operations of an organization and are included in application systems that automate key business processes in different areas such as sales, service, banking, order management, manufacturing, aviation, purchasing, billing, etc. Some examples of transaction data 232 include enterprise data (e.g. order-entry, supply-chain, shipping, invoices), sales data (e.g. accounts, leads, opportunities), aviation data (carriers, bookings, revenue), and the like.
Most often, the integration process includes accumulating transaction data of a different format than what is ultimately needed for analytic operations. The process of acquiring transaction data and converting it into useful, compatible and accurate data can include three, or more, phases such as extract, load and transform. In some implementations, the integration flow can include various integration flow styles. One such style can be Extract-Transform-Load (ETL), where, after extraction from a data source, data can be transformed and then loaded into a data warehouse. In another implementation, an Extract-Load-Transform (ELT) style can be employed, where, after the extraction, data can be first loaded to the data warehouse and then transformation operation can be applied. In yet another implementation, the integration can use an Extract-Transform-Load-Transform (ETLT) style, where, after the extraction, several data optimization techniques (e.g. clustering, normalization, denormalization) can be applied, then the data can be loaded to the data warehouse and then more heavy transformation operations can occur.
Extraction refers to the task of acquiring transaction data from transactional data stores, according to one implementation. This can be as simple as downloading a flat file from a database or a spreadsheet, or as sophisticated as setting up relationships with external systems that then control the transportation of data to the target system. Loading is the phase in which the captured data is deposited into a new data store such as a warehouse or a mart. In some implementations, loading can be accomplished by custom programming commands such as IMPORT in structured query language (SQL) and LOAD in Oracle Utilities. In some implementations, a plurality of application-programming interfaces (APIs) can be used, to interface with a plurality of transactional data sources, along with extraction connectors that load the transaction data into dedicated data stores.
Transformation refers to the stage of applying a series of rules or functions to the extracted or the loaded data, generally so as to convert the extracted or the loaded data to a format that is conducive for deriving analytics. Some examples of transformation include selecting only certain columns to load, translating coded values, encoding free-form values, deriving new calculated values, sorting, joining data from multiple sources, aggregation, de-normalization, transposing or pivoting data, splitting a column into multiple columns and data validation.
In some implementations, ELT workflow generates a so-called precursor edgemart by performing lightweight transformations on the transaction data. One example of a light-weight transformation is denormalization transformation. A denormalization transformation reintroduces some number of redundancies that existed prior to normalization of the transaction data 232, according to one implementation. For instance, a denormalization transformation can remove certain joins between two tables. The resulting so-called precursor edgemart has lesser degrees of normal norms relative to the transaction data, and thus is more optimum for analytics operations such as faster retrieval access, multidimensional indexing and caching and automated computation of higher level aggregates of the transaction data.
In other implementations, the loaded data can undergo a plurality of heavy-weight transformations, including joining data from two related edgemart(s), flattening the transaction role hierarchy to enable role-based security, increasing query performance on specific data and registering an edgemart to make it available for queries. Depending on the type of transformation, the data in an existing edgemart is updated or a new edgemart is generated.
In one implementation of the heavy-weight transformations, an augment transformation joins data from two edgemart(s) to enable queries across both of them. For instance, augmenting a “User EdgeMart” with an “Account EdgeMart” can enable a data analyst to generate query that displays all account details, including the names of the account owner and creator. Augmentation transformation creates a new edgemart based on data from two input edgemart(s). Each input edgemart can be identified as the left or right edgemart. The new edgemart includes all the columns of the left edgemart and appends only the specified columns from the right edgemart. Augmentation transformation performs a left, outer join, where the new edgemart includes all rows from the left edgemart and only matched rows from the right edgemart. In another implementation, queries can be enabled that span more than two edgemart(s). This can be achieved by augmenting two edgemart(s) at a time. For example, to augment three edgemart(s), a first two edgemart(s) can be augmented before augmenting the resulting edgemart with a third edgemart.
In some implementations, a join condition in the augment transformation can be specified to determine how to match rows in the right edgemart to those in the left edgemart. The following example illustrates a single-column join condition. To augment the following edgemarts based on single-column key, an “Opportunity” is assigned as the left edgemart and an “Account” is assigned as the right edgemart. Also, “OpptyAcct” is specified as the relationship between them.
Upon running an ELT workflow job, an “OpptyAcct” prefix is added to all account columns and the edgemarts are joined based on a key defined as “Opportunity.Account_ID=Account.ID.” After running the ELT workflow job to augment the two input edgemarts, the resulting edgemart includes the following columns:
In other implementations, different heavy-weight transformations can be applied, including flatten transformation to create role-based access on accounts, index transformation to index one dimension column in an edgemart, Ngram transformation to generate case-sensitive, full-text index based on data in an edgemart, register transformation to register an edgemart to make it available for queries and extract transformation to extract data from fields of a data object.
In one implementation, an interface system 202 implementing a load balancing function (e.g., an F5 Big-IP load balancer) is communicably coupled between the servers 514 and the superpod engine 204 to distribute requests to the worker servers 528. In one aspect, the load balancer uses at least virtual IP (VIP) templates and connections algorithm to route user requests to the worker servers 528. A VIP template contains load balancer-related configuration settings for a specific type of network traffic. Other examples of load balancing algorithms, such as round robin and observed response time, also can be used. For example, in certain aspects, three consecutive requests from the same user could hit three different worker servers, and three requests from different users could hit the same worker server. In this manner, transactional database management system 232 is multi-tenant, wherein integration environment handles storage of, and access to, different objects, data and applications across disparate users and organizations.
Superpod engines 204 also host the queuing engine 208, which in turn implements a key-value server 518 that is in communication with a key-value store. Key-value store is a type of storage that enables users to store and read data (values) with a unique key. In some implementations, a key-value store stores a schema-less data. This data can consist of a string that represents the key and the actual data is the value in the “key-value” relationship. According to one implementation, the data itself can be any type of primitive of the programming language such as a string, an integer, or an array. In another implementation, it can be an object that binds to the key-value store. Using a key-value store replaces the need of fixed data model and makes the requirement for properly formatted data less strict. Some popular examples of different key-value stores include Redis, CouchDB, Tokyo Cabinet and Cassandra. The example shown in
In some implementations, queuing engine 208 sets server affinity for a user and/or organization to a specific work server 528 or to a cluster of worker servers 528. Server affinity refers to the set up that a server or servers in a same cluster are dedicated to service requests from the same client, according to one implementation. In another implementation, server affinity within a cluster of servers refers to the set up that when a server in the cluster fails to process a request, then the request can only be picked by another server in the cluster. Server affinity can be achieved by configuring the load balancers 202 such that they are forced to send requests from a particular client only to corresponding servers dedicated to the particular client. Affinity relationships between clients and servers or server clusters are mapped in a directory service. Directory service defines a client name and sets it to an IP address of a server. When a client name is affinitized to multiple servers, client affinity is established once a request's destination IP address matches the cluster's global IP address.
As described above, an edgemart 142 is a collection of data optimized for queries. An edgemart 142 can be one file comprising a plurality of dimensions and measures, or a set of tables, each comprising a plurality of dimensions and measures. These queries that are applied to the edgemart 142 are referred to as lenses. A lens is a predefined view (query) that can be run against one or more edgemarts 142. The lens can be as simple as the default explorer lens on the edgemart 142 (count of rows) or as complex as an interactive dashboard (surfacing multiple edgemart views and interactive filters). Lenses can be stored for reuse, or can be created ad hoc for execution. Lenses can also be scheduled for execution on a periodic basis. The result of the lens execution on the edgemart 142 can be an export of data to be presented to a user device(s) 255, or can result in a split edgemart 216, also known as a shard.
Lens can be expressed as a URL, which can trigger a high performance server side query using EQL (a derivative of Pig Latin), according to some implementations. In other implementations, less demanding queries can also be performed on the edgemart 142 via a JavaScript query engine after the edgemart 142 has been copied to local storage coupled to the user device(s) 255.
Server-independent availability of the edgemart 142 allows for offline processing of the locally storage edgemarts (s), thus providing users with a larger window of operation. Also, local storage reduces the server load and the query processing time. In addition, it allows the service providers to shift certain degree of onus of processing on respective client devices and thus save on extensive cost associated with infrastructure installation and maintenance.
The technology disclosed uses a plurality of predictive pre-fetching and caching strategies to improve response time of queries issued using portable mobile devices. In particular, the technology disclosed performs anticipatory migration of dimensions and measures to local storages coupled to the portable mobile devices responsive to user needs. In one implementation, this is achieved by determining patterns of dimensions and measures requested by a user or groups of users over a time period. In another implementation, this is achieved by recognizing dimensions and measures used to generate query results for the user or the groups of users over a time period. In yet another implementation, this is achieved by identifying dimensions and measures accessible to the user or the groups of users based on their roles, assigned group, user ID, location (geolocation, region, time zone), etc.
The technology disclosed further enhances user experience by automating the decision of whether the data extract queries are executed against an edgemart 142 stored on a server, or against a local copy of the edgemart 142 already copied to the local storage because of its practical size and volume. This is especially interesting when graphic data animation is involved because of the complexity of the animation. In the animation context, locally stored rich animation data can be efficiently used instead of server side copy of the same, to which a substantial migration time is attached.
Edgemart(s) 142 are created from queries against transaction data 232. When the queries require transformation of data, the process is referred to as digestion, and is part of the ELT workflow described above. The Insights® edgemart engine 152 is the tool used to create and manage edgemart(s) 142. For example, in one implementation, Insights® is installed with a Salseforce.com® pod engine 222 with its associated transaction data 232, and is used as a data source for an edgemart 142.
In another implementation, a user device 255 uses the edgemart builder UI within the explorer engine 102 to modify a default digestion workflow on a pod engine 222, e.g., to exclude non-analytical custom fields from Accounts/Opportunities tables within a transaction data set 232 and to include a custom object related to Opportunity. The digestion workflow is scheduled on an edgemart engine 152. When triggered, the digestion workflow first queries the Salesforce API in the pod engine 222 to fetch the relevant data and metadata, converts the data into a set of raw edgemarts, and performs necessary transformations to produce the final, customer visible edgemarts 142. The final edgemarts are stored with the transaction data 232 for persistence and disaster recoverability.
Further, the user device 255 is notified that the edgemart is available. The user device 255 accesses the explorer lens for the edgemart 142 to dynamically explore the data. If the edgemart is large, the technology disclosed chooses to execute the lens using EQL against the edgemart 142 stored with the transaction data 232. Alternatively, the technology disclosed can choose to copy the edgemart 142 stored with the transaction data 232 to the user's device 255, and execute the subsequent lens locally on the user's device 255.
This evaluation includes taking into account:
In one implementation, a so-called EQL language is used to process edgemarts. EQL language is a real-time query language that uses data flow as a means of aligning results. It enables ad hoc analysis of data that is stored in edgemarts. An EQL script consists of a sequence of statements that are made up of keywords (such as filter, group, and order), identifiers, literals, or special characters.
In some implementations, if the edgemart is considered small, then the edgemart is copied to the local storage 702, such as local edgemarts 252, where all queries are performed against the local copy of the edgemart. This can greatly improve the performance of the queries, and is referred to as edge computing. Edge computing (including query and digest/transform) can use a JavaScript that supports interactive querying within a browser/mobile client and also supports offline/disconnected queries.
In other implementations, the edgemart evaluation 632 can be overridden such that an edgemart can be classified as large regardless of its attributes. This allows a customer to choose all queries to be executed against the edgemart stored in the server storage 710 regardless of the amount of data in the edgemart, the processing capabilities of the user device, or the available bandwidth between the user device and the server storage.
At exchange 612, user device 604, such as user device(s) 255, requests authentication with a security server 606 such as security engine 245.
At exchange 622, security server 606 authorizes the user device 604 by providing a security token 622 to the user device 604.
At exchange 636, user device 604 issues a first query to the application server 608, which requires creation of edgemart(s) and presentation of data from the edgemart(s) across the user device 604.
At exchange 646, application server 608 provides the user device 604 with edgemart data 646 stored in the server storage 610. In one implementation, this is achieved by first sending the edgemart data to a local storage 602, such as local edgemarts 252, at exchange 646 and then forwarding the edgemart data to the user device 604 at exchange 648. In some implementations local storage 602 is part of the user device 604 in the form of non-volatile memory unit. In other implementations, local storage 602 is coupled to the user device 604.
At exchange 656, user device 604 issues additional queries to the application server 608. In response, application server 608 performs an evaluation 658 that whether enough complete fields of the edgemart data have been migrated to the local storage 602 at exchange 646. If enough complete fields of the edgemart data have not been migrated to local storage 602 at exchange 646, then the additional queries are run on the server storage 610 at exchange 659 and the retrieved data is sent to the user device 604 at exchange 682 via exchange 679.
To the contrary, if enough complete fields of the edgemart data have been migrated to local storage 602 at exchange 646, then additional queries are run on the local storage 610 at exchange 689 and the retrieved data is sent to the user device 604 at exchange 692.
At exchange 712, user device 704, such as user device(s) 255, requests authentication with a security server 706 such as security engine 245.
At exchange 722, security server 706 authorizes the user device 704 by providing a security token 722 to the user device 704.
At exchange 736, user device 704 issues a first query to the application server 708, which requires creation of edgemart(s) and presentation of data from the edgemart(s) across the user device 704.
In response, application server 708 performs an edgemart evaluation 727 that whether the size of the edgemart required for processing the first query is large or small. In this context, the term large means that the edgemart evaluation 727, based on the amount of data in the edgemart, the processing capabilities of the user device 704, or the available bandwidth between the user device 704 and the server storage 710, calculates that the edgemart cannot be transmitted to the user device 704 for the execution of a local query in a timely manner. In this case, all query processing is executed on the server storage 710 at exchange 728 and only the results are communicated to the user device 704. In one implementation, results are communicated to the user device 704 by first sending the edgemart data to a local storage 702 at exchange 738 and then forwarding the edgemart data to the user device 704 at exchange 752. In some implementations local storage 702 is part of the user device 704 in the form of non-volatile memory unit. In other implementations, local storage 702 is coupled to the user device 704. In one implementation, a value corresponding to a large edgemart is user-definable such that a size specified by a user, such as 510 MB or 1 GB, is considered a lower limit for small edgemart.
Further, the term small means that the edgemart evaluation 727, based on the amount of data in the edgemart, the processing capabilities of the user device 704, or the available bandwidth between the user device 704 and the server storage 710, calculates that the edgemart can be transmitted to the user device 704 for the local execution of a query in a timely manner at exchange 758 and presentation to the user device 704 at exchange 762. In one implementation, a value corresponding to a small edgemart is user-definable such that a size specified by a user, such as 510 MB or 1 GB, is considered an upper limit for small edgemart.
At exchange 762, user device 704 issues additional queries to the application server 708. In response, application server 708 performs an evaluation 764 that whether enough complete fields of the edgemart data have been migrated to the local storage 702 at exchange 738 and/or exchange 758. If enough complete fields of the edgemart data have not been migrated to local storage 702 at exchange 738 and/or exchange 758, then the additional queries are run on the server storage 710 at exchange 768 and the retrieved data is sent to the user device 704 at exchange 762 via exchange 778.
To the contrary, if enough complete fields of the edgemart data have been migrated to local storage 702 at exchange 738 and/or exchange 758, then additional queries are run on the local storage 710 at exchange 788 and the retrieved data is sent to the user device 704 at exchange 792.
At action 802, a first query is initiated to a server against a database.
At action 812, first query results are received from the server.
At action 822, parts of the database are received on a field-by-field basis from the server and the parts are retained locally. In one implementation, the parts of the database are received without actively requesting the parts. In another implementation, the parts of the database are received in an order that correlates with fields used to construct dashboards most recently or frequently used by a user. In some implementations, at least some of the parts include an index of a dimension and data in the dimension are not repeated in a corresponding data record that is retained locally. In other implementations, at least some of the parts include a measures field of data subject to aggregation.
In one implementation, the server is sent a user identifier that links the user to a user history of queries against particular fields and the parts are received in an order positively correlated with a set of the queries in the user history.
In another implementation, the server is sent a user identifier that links the user to one or more dashboards that use particular fields and the parts are received in an order positively correlated with fields in the dashboard.
In yet another implementation, the server is sent a user identifier that links the user to at least one of a role, group, and location that specify access to particular data rows, records or objects.
At action 832, a plurality of additional queries are initiated and for each additional query a determination is made that whether enough complete fields have been received to locally process to the additional queries without further querying the server.
At action 842, responsive to the determination, either the additional queries are locally processed or sent to the server for processing. In one implementation, the parts of the database used to locally process the additional queries are all received from the server after initiating the first query. In one implementation, visiting a home page of the database causes any of the parts previously received and retained locally to be invalidated.
This method and other implementations of the technology disclosed can include one or more of the following features and/or features described in connection with additional methods disclosed. In the interest of conciseness, the combinations of features disclosed in this application are not individually enumerated and are not repeated with each base set of features. The reader will understand how features identified in this section can readily be combined with sets of base features identified as implementations in sections of this application such as analytics environment, integration environment, ELT workflow, integration components, offloading search processing, client side search experience, etc.
Other implementations may include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation may include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.
At action 902, program instructions executable on a client that, for each additional query after a first query to a server against a database, cause the client to determine whether enough complete fields of the database have been received to locally process to the additional queries without further querying the server.
At action 912, responsive to the determination, either the additional queries are processed locally or the additional queries sent to the server, but not both.
At action 922, the first query is received from the client.
At action 932, first query results are transmitted to the client.
At action 942, parts of the database are transmitted on a field-by-field basis to the client to be retained locally and used by the program instructions to locally process additional queries. In yet another implementation, the parts of the database used by the client to locally process the additional queries are all transmitted to the client after receiving the first query. In some implementations, at least some of the parts include an index of a dimension and data in the dimension are not repeated in a corresponding data record that is retained locally. In other implementations, at least some of the parts include measures field of data subject to aggregation.
In one implementation, the parts of the database are transmitted in an order that correlates with fields used to construct dashboards most recently or frequently used by a user identified with the first query, without receiving an active request for the parts. In another implementation, a field priority request is received from the client for at least some of the parts of the database and he parts transmitted are ordered at least in part on the field priority request.
In one implementation, a user identifier associated with the first request is received, a corresponding user history of queries against particular fields is accessed and the parts are transmitted in an order positively correlated with a set of the queries in the user history.
In another implementation, a user identifier associated with the first request is received, a list of one or more dashboards that use particular fields is accessed and the parts are transmitted in an order positively correlated with the particular fields used to construct the dashboards.
In yet another implementation, a user identifier is received that links the user to a role and the parts are transmitted in an order based at least in part on the role.
In a further implementation, a user identifier is received that links the user to a role that has access to particular data rows, records or objects and data in the parts transmitted is limited based on data access right of the role.
In one implementation, web page requests by the client are tracked and, upon receiving a request for a home page of the database, any of the parts previously transmitted are queued for re-transmission.
This method and other implementations of the technology disclosed can include one or more of the following features and/or features described in connection with additional methods disclosed.
Other implementations may include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation may include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.
At action 1002, a first query is initiated to a server against a database.
At action 1012, first query results are received from the server and the first query results are locally retained on a field-by-field basis dependent on at least one of size of the first query results, available bandwidth and available computing power.
At action 1022, a plurality of additional queries are initiated and for each additional query a determination is made that whether enough complete fields have been received to locally process to the additional queries without further querying the server.
At action 1032, responsive to the determination, either the additional queries are locally processed or sent to the server for processing. In one implementation, the parts of the database used to locally process the additional queries are all received from the server after initiating the first query. In one implementation, visiting a home page of the database causes any of the parts previously received and retained locally to be invalidated.
This method and other implementations of the technology disclosed can include one or more of the following features and/or features described in connection with additional methods disclosed.
Other implementations may include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation may include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.
User interface input devices 1122 can include a keyboard; pointing devices such as a mouse, trackball, touchpad, or graphics tablet; a scanner; a touch screen incorporated into the display; audio input devices such as voice recognition systems and microphones; and other types of input devices. In general, use of the term “input device” is intended to include all possible types of devices and ways to input information into computer system 1110.
User interface output devices 1118 can include a display subsystem, a printer, a fax machine, or non-visual displays such as audio output devices. The display subsystem can include a cathode ray tube (CRT), a flat-panel device such as a liquid crystal display (LCD), a projection device, or some other mechanism for creating a visible image. The display subsystem can also provide a non-visual display such as audio output devices. In general, use of the term “output device” is intended to include all possible types of devices and ways to output information from computer system 1110 to the user or to another machine or computer system.
Storage subsystem 1124 stores programming and data constructs that provide the functionality of some or all of the modules and methods described herein. These software modules are generally executed by processor 1114 alone or in combination with other processors.
Memory 1126 used in the storage subsystem can include a number of memories including a main random access memory (RAM) 1130 for storage of instructions and data during program execution and a read only memory (ROM) 1132 in which fixed instructions are stored. A file storage subsystem 1128 can provide persistent storage for program and data files, and can include a hard disk drive, a floppy disk drive along with associated removable media, a CD-ROM drive, an optical drive, or removable media cartridges. The modules implementing the functionality of certain implementations can be stored by file storage subsystem 1128 in the storage subsystem 1124, or in other machines accessible by the processor.
Bus subsystem 1112 provides a mechanism for letting the various components and subsystems of computer system 1110 communicate with each other as intended. Although bus subsystem 1112 is shown schematically as a single bus, alternative implementations of the bus subsystem can use multiple busses. Application server 1120 can be a framework that allows the applications of computer system 1110 to run, such as the hardware and/or software, e.g., the operating system.
Computer system 1110 can be of varying types including a workstation, server, computing cluster, blade server, server farm, or any other data processing system or computing device. Due to the ever-changing nature of computers and networks, the description of computer system 1110 depicted in
The terms and expressions employed herein are used as terms and expressions of description and not of limitation, and there is no intention, in the use of such terms and expressions, of excluding any equivalents of the features shown and described or portions thereof. In addition, having described certain implementations of the technology disclosed, it will be apparent to those of ordinary skill in the art that other implementations incorporating the concepts disclosed herein can be used without departing from the spirit and scope of the technology disclosed. Accordingly, the described implementations are to be considered in all respects as only illustrative and not restrictive.
