Patent Application
20050091215
1. Field of the Invention
The present invention relates to data storage systems. More particularly, the present invention relates to a system and a method for allocating and de-allocating servers in a Storage Area Network (SAN).
2. Description of the Related Art
Servers in Intranet and Internet data-centers, commonly referred to as on-demand servers, are significantly over-provisioned and, consequently, have a relatively low utilization rate. One conventional solution for managing storage in an on-demand server environment having multiple servers is to associate dedicated disks (local disks or Storage Area Network (SAN)—attached disks) with each respective server. All of the state data that is required for a server is copied onto a disk at the time that the disk is dedicated to the server and added to a cluster of dedicated disks associated with a server. A significant amount of time may be required to copy the state data onto the newly added disk. Accordingly, the total time required to allocate a server to a system user can be significant when a conventional dedicated-disk-and-an-all-state-data-copy allocation technique is used. As used herein, the term “system user” means a customer and/or an application.
Additionally, each dedicated disk must be scrubbed, that is, state data associated with a system user removed, when the server is de-allocated to avoid a leak of confidential data between system users. Consequently, a further increase in time is required when a dedicated disk is de-allocated. The large amounts of time that are required for allocation and de-allocation of a server results in a significant under-utilization of resources. See, for example, A. Chandra et al., “Quantifying the benefits of resource multiplexing in on-demand data centers,” Proceedings of the First ACM Workshop on Algorithms and Architectures for Self-Managing Systems (Self-Manage 2003), June 2003. Thus, a conventional dedicated-disk-and-an-all-state copy allocation technique is desirable only when the amount of persistent state data is small.
Another conventional allocation and de-allocation technique that is used for some on-demand server systems is a Network File System (NFS) server for the entire state data. Although such a technique reduces the overall time required for allocation and de-allocation, there are, nevertheless, disadvantages. For example, NFS performance is lower than the performance of a local file system. Additionally, a NFS server can not be conveniently shared among multiple system users because there are security issues, as well as potential user identification (userid) conflicts.
Consequently, what is needed is an allocation and de-allocation technique for an on-demand system that reduces the time required for allocating and de-allocating servers for a system user.
The present invention provides an allocation and de-allocation technique for an on-demand system that reduces the time required for allocating and de-allocating servers for a system user.
The advantages of the present invention are provided by a system and a method of allocating storage to a system user, such as a customer or an application, of a storage area network that includes storage and a plurality of servers accessing the storage. According to the invention, at least one master storage image that is in the storage and that will be associated with a system user when a server is allocated to the system user is identified. The master storage image is pre-configured with data and state information that is associated with a system user. A plurality of replicas of each identified master storage image is generated prior to at least one server being allocated to the system user. Each replica is a logical volume. A selected replica of the plurality of replicas is allocated to each server allocated to the system user. Each time a server is de-allocated, the corresponding replica is de-allocated and assigned to a pool of de-allocated replicas. The pool of de-allocated replicas is configured to automatically scrub all replicas, such as by reformatting, asynchronously when, for example, the number of de-allocated replicas assigned to the pool equals a predetermined number.
The present invention also provides a system and a method for allocating storage between system users of a storage area network that includes storage and a plurality of servers accessing the storage. At least one master storage image that is stored in the storage of the storage area network and that will be associated with a system user is identified. Each master storage image includes both a read-only data portion and a writeable data portion. A read-only copy of the read-only data portion of each master storage image is generated and shared across the plurality of servers. The read-only copy of the read-only data portion of a selected master storage image is allocated to each server that is allocated to the system user. A separate writable data volume of the writable data portion of the selected master storage image is allocated to each server allocated to the system user. When a server that has been allocated to the system user is de-allocated, the read-only copy of the read-only data portion of the selected master image is de-allocated from the de-allocated server. The writable data volume allocated to the server that has been de-allocated is de-allocated and assigned to a pool of de-allocated writable data volumes that is automatically scrubbed, such as by reformatting, asynchronously from the de-allocation of the writable data volume when, for example, the amount of writable data volumes assigned to the pool equals a predetermined amount.
The present invention is illustrated by way of example and not by limitation in the accompanying figures in which like reference numerals indicate similar elements and in which:
The present invention increases server utilization for an on-demand system by dynamically sharing servers between multiple system users. The number of servers and the amount of storage allocated to a particular system user is dynamically changed based on the traffic received by the on-demand system.
The following description of the present invention will use web/application servers as exemplary servers because web/application servers have two types of persistent state data, mostly-read-only state data, such as html files, and mostly-write state data that is local to the server, such as logs. It should be kept in mind, though, that the servers that are dynamically allocated by the present invention may be used for a variety of purposes, such as web servers, application servers, databases, etc., and are not limited to the exemplary web/application servers that are described herein.
All servers 101a-101d are dynamically allocated by virtue of being connected to on-demand server system 100, which is configured as a SAN. In a conventional SAN environment, every server can undesirably access every logical drive in the SAN environment unless additional measures are taken. In contrast to a conventional SAN environment, the present invention restricts a system user so that each system user can access only the particular set of logical drives that have been allocated to the system user. Inter- and intra-system user security and access controls are dynamically implemented by SAN-attached storage controllers 104a and 104b that use SAN Logical Unit Number (LUN) masking and SAN zoning. That is, SAN-attached storage controllers 104a and 104b restrict system user access using LUN masking to a set of logical drives corresponding to servers allocated to the system user. SAN zoning provides an additional level of security by creating virtual networks for each system user so that only the ports and devices that are required to access the logical drives allocated to the system user belong to the specified zone for the system user.
To reduce the time necessary for allocating and de-allocating web/application servers to a system user, the present invention utilizes a pre-configuration technique by creating a predetermined number X of replicas 106a-106d of a master storage image 107 for a system user, such that each replica is a logical drive within the SAN. A master storage image can contain, for example, a boot image, application code and/or data, web pages, etc. More than one replica can be stored on a storage unit because each storage unit can contain more than one logical drive. When a new server is allocated to a system user, one of the replicas is also allocated to the server. The logical drive containing the allocated replica serves as the local hard disk for the system user on which all persistent application state data is stored. When a server is de-allocated from the system user, each logical drive that has been allocated to the system user is de-allocated and added to a pool of logical drives that must be scrubbed. For example, let Y be the maximum number of disks in the pool that must be scrubbed. By appropriately selecting a number X of replicas for each user and a number Y of disks in the pool that must be scrubbed for the overall system, a server is not required to wait for (1) data to be copied when the server is being allocated to a system user or (2) allocated disk to be scrubbed when the server is being de-allocated, thereby desirably reducing the total time of the allocation and de-allocation process.
A drawback that is associated with pre-configuring X replicas for a system user is that more disks are used than are strictly necessary. To overcome this drawback, the present invention reduces the disk space requirement by exploiting the fact that a significant amount of data is read-only data. The present invention separates the persistent application state data into mostly read-only state data and read-write state data. The read-only state data is stored on a single logical drive that is shared by all servers. All of the servers mount the logical drive having read-only state data as a read-only file system. The read-write state data that is local to each server is stored on a separate logical drive for each server and is de-allocated and scrubbed when the server is de-allocated. Zoning keeps one system user from accessing read-only state data and writable state data of another system user even though a single logical drive is shared by multiple system users.
Read-only data sharing according to the present invention is as fast as a conventional pre-configuration technique for allocation and de-allocation, but requires significantly less disk space than a conventional pre-configuration technique. In that regard, the disk-space requirement of the read-only state data sharing technique of the present invention is comparable to disk-space requirement of a conventional Network-Attached Storage device (NAS) technique.
The read-only state data sharing technique of the present invention requires a special technique for updating the read-only state data, such as when html files in a web-server are updated. To handle update of read-only state data, the new read-only state data is created on a new logical drive, that is, a logical drive that has not yet been allocated. The currently allocated read-only logical drive is then changed to the new logical drive by rebooting the servers belonging to a cluster one-at-a-time, thereby ensuring that the cluster for the server is always up and with at most one server being down at any time.
At step 301 in
Although the foregoing invention has been described in some detail for purposes of clarity of understanding, it will be apparent that certain changes and modifications may be practiced that are within the scope of the appended claims. Accordingly, the present embodiments are to be considered as illustrative and not restrictive, and the invention is not to be limited to the details given herein, but may be modified within the scope and equivalents of the appended claims.
