Matching application requirements with appropriate storage resources enables administrators to fully realize the value of networked storage.
By Bill Pierce
Business applications will continue to demand increasing amounts of storage. Applications driving capacity growth include e-mail, e-commerce, transactional databases, digital photography/audio/video, medical imaging, and regulation-required documentation (e.g., the Health Insurance & Portability Accounting Act [HIPAA]).
Not all storage solutions are created equal. Each application requires storage that has specific characteristics in the areas of capacity, performance, security, and availability. Purchasing the biggest, newest, and fastest storage components is not only economically unfeasible, but it is also impractical when such characteristics are unnecessary. Instead, legacy equipment should be fully leveraged for maximum return on investment.
Storage investment can be extended through intelligent distribution of content into repositories having those characteristics necessary to meet application needs. This "boutique storage" approach tailors optimal storage characteristics to match individual business application requirements.
Storage area network (SAN) and network-attached storage (NAS) technologies provide flexible options for application-driven storage, maintaining service levels and providing economies of scale. However, this increased flexibility adds complexity. SAN/NAS topologies can route disparate application traffic across multiple device types, systems interfaces, and data-transfer protocols, making it increasingly difficult to track which applications are using what storage and to know whether applications and storage devices are appropriately matched.
Role of SRM in SAN/NAS
Storage resource management (SRM) software can ease much of the complexity of implementing, administering, and optimizing SAN and NAS environments by automating analysis of current application demands and relative storage use. Customization of service level agreements can provide the ability to meet actual needs and enables tailoring of departmental chargeback fees to storage use. Just-in-time (JIT) deferral of storage subsystem purchases can be accomplished through a more accurate projection of application needs. Accordingly, SRM can substantially decrease staff hours spent in storage network planning, configuration, implementation, administration, and maintenance.
By facilitating appropriate storage access and allocation throughout the SAN/NAS infrastructure, SRM can leverage sunk cost in legacy storage components and streamline overall system efficiency. Cumulatively, SRM can significantly reduce the total cost of owning and maintaining storage resources.
According to industry analysts, by applying storage management tools like SRM, IT management will be able to increase disk-space usage from existing ranges of 30% to 60% to as much as 85% to 90%.
Matching applications and storage
SAN and NAS environments comprise a complex chain of components, or storage layers (see table on the left). Each layer has unique characteristics that affect the overall appropriateness of application storage. SRM allows administrators to coordinate these characteristics into a unified application-storage solution.
To appropriately match applications to storage, specific application requirements and storage characteristics must be identified. Applications can be grouped by their primary need for space, performance, availability, security, and accessibility (see table, above). In each case, administrative requirements of flexibility, simplicity, cost, and priority also must be weighed.
Space-driven applications consume vast and ever-growing amounts of storage. They must be provisioned to scalable platforms containing abundant available capacity and should also employ technologies that incur minimal disruption when expansion requires relocation or reconfiguration of storage volumes.
High-capacity requirements may be best met with SAN-based RAID arrays that allow expansion of an existing LUN by extension from their own storage pool. SAN-based virtualization schemes can provision LUNs across RAID array boundaries. Volume managers can deliver additional redirection flexibility. Resizable file systems (e.g., Linux ext2) support space-driven, file-based applications by permitting file system resizing after resident volumes have been expanded.
For performance-driven applications, time spent blocking application threads during I/O transfers must be minimized. To fine-tune a storage solution for performance-driven applications, additional application requirements must be specified (e.g., latency [in applications with primarily small I/O transfers], throughput [in applications with primarily large I/O transfers], expected read/write ratios, and repetitiveness of data access [for cache design]).
Each of the storage layers affects overall performance. Because of lower protocol overhead and greater potential for parallel data access, a SAN generally provides greater performance than NAS. Throughput may be increased through striping in the volume management, device driver, virtualization, and storage device layers. By avoiding parity checks, RAID 0 achieves better performance than RAID 5 but at the cost of reduced reliability. Caching in the operating system, virtualization, and storage device layers can reduce latency. Unnecessary file system journaling also can affect performance by increasing overhead.
Availability-driven applications demand business continuance in the event of catastrophic failures. Mirroring and fail-over in the volume manager, driver, virtualization, and storage device layers can provide this redundancy. In providing redundant data paths, SANs are typically more flexible than NAS. Data can also be mirrored or replicated to distant devices via Fibre Channel-over-IP links to provide availability when physical disruption occurs at one site.
Where performance is also important, parity-checking RAID (RAID 3 or 5) is better than RAID 0 because it combines striping with redundancy. Storage arrays that use hot-spare technology can significantly improve mean time to data loss by automatically replacing and rebuilding failed drives. Storage arrays and NAS devices that provide point-in-time copies can recover faster than archival tape backup schemes. Journaling file systems can speed recovery in the event of file system corruption.
Security-driven applications can demand restriction of data access and protection from data loss. For access restriction, physical security is important. Direct-attached storage (DAS) is more secure than SAN or NAS, which expose access points to data streams and repositories. Because of inherent access permissions and access-auditing capability, the NTFS file system on Windows 2000 is preferable to a FAT32 file system.
High-availability technologies and redundancy can help prevent data loss, and journaling file systems help prevent data corruption. Volatile caches improve performance, but present a risk of potential data loss in the event of power failure. Specialized backup policies (e.g., full daily backup to direct-attached tape drives) can also increase data recoverability.
Accessibility-driven applications must be shared safely by a wide range of users. NAS is inherently more accessible than SAN storage, providing easy network access to multiple platforms through a variety of protocols, including NFS, CIFS, HTTP, and FTP.
SRM software can alert administrators about performance bottlenecks and recommend capacity re-allocation to increase performance.
Each protocol offers its own level of security, with authentication and authorization that is compatible with established schemes such as NIS+, Active Directory, and LDAP. Some NAS filers offer file-locking mechanisms that allow synchronization of concurrent read-write access at the file level. In contrast, SAN storage requires specialized clustering software to allow the same storage to be concurrently shared safely.
In practice, application requirements must be balanced with the administrative considerations of flexibility, simplicity, cost, and business priorities. Storage flexibility is necessary when an application has multiple requirements (e.g., availability and security). Simple solutions are important where staff is limited or schedules compressed. Lower-cost alternatives must be examined if the best storage choice does not fit within budget. SRM tools are indispensable for effectively achieving balanced storage solutions.
Optimizing SAN/NAS with SRM
SRM can maximize the value of SAN and NAS environments in all phases of storage management: assessment, provisioning, monitoring, planning, and configuration.
Storage assessment—The storage assessment phase consists of determining 1) those storage resources already in place; 2) how applications are currently using them, by displaying used and unused storage space and other qualities.
Examination of application content residing on storage space and measurement of how the application drives the storage layers can determine whether storage resources are optimized to meet demand. By re-allocating unused legacy storage, SRM can reduce storage TCO by leveraging sunk costs. Furthermore, by correlating data ownership with underlying storage, SRM facilitates chargeback, closing the loop of storage-resource utilization and cost. SRM tools can view and analyze application use of storage space, as well as bandwidth, content ownership, and the qualities of underlying storage components.
Assessment scenario—SRM discovers that a NAS filer is concurrently supporting a read-intensive Oracle transaction database and a vast array of project directories, causing a performance bottleneck on the LAN. SRM software alerts management to the problem and recommends re-allocation of 2TB of otherwise unused RAID 5 space available on a disk array on the SAN, diverting traffic from the LAN and increasing database throughput (see figure, p. 34).
Provisioning—Provisioning prioritizes application needs and allocates existing capacity identified during the assessment phase. SRM software can intelligently search current resources based on pre-defined application needs and return a weighted list of potential storage options, similar to "relevance-weighted" Internet searches. Using SRM in SAN/NAS environments, resources can be optimized as they are consumed.
Provisioning scenario—A 500MB directory is required to store highly sensitive strategic plans. Only certain Windows XP workstations, connected to a file server behind an internal firewall, can access the files. SRM locates an underutilized 1GB partition of DAS on a Windows 2000 Server that is securely within the same domain.
Monitoring—Storage usage is dynamic. Existing storage resources are frequently stretched to meet increasing application demands. To avoid outages, limits can be defined to trigger manual, automated, or automation-assisted responses. Conditions underlying service levels can shift, invalidating prior storage optimizations. SRM can monitor the SAN/NAS environment to track these shifting trends and confirm the ongoing validity of these optimizations, providing timely notification to identify storage issues before they become problems.
Monitoring scenario—A bioinformatics application with high-performance requirements was allocated certain LUNs through a controller on a disk array. Other applications have since been provisioned to this same channel. As demand on the resource has grown, available bandwidth has degraded. Without SRM, unacceptably long run times can develop. Monitoring I/O performance, SRM software sends alerts before implicit service levels become compromised. It identifies other applications contending for that bandwidth and helps redirect them to available SAN/NAS pool resources.
Planning—New storage-resource acquisition requires lead time for budgeting, selection, implementation, and integration. Economically, expenditures on new storage must be delayed until absolutely required, yet before service degradation or outages occur. In DAS environments, planning storage allocation by individual system can result in the purchase of inefficient, small storage systems, suited only to the needs of individual servers.
SRM allows management of the entire SAN/NAS environment as one consolidated resource pool. Consumption of storage with different characteristics may occur at different rates. SRM can monitor multiple network storage types, relative rates of consumption, and underlying storage characteristics, facilitating acquisition planning.
Planning scenario—SRM trending shows NAS storage use is growing at an average rate of 600GB per quarter—far faster than the growth of SAN-based RAID. At this rate, filer space will be consumed in two quarters. SRM can be configured to issue quarterly alerts to acquire more NAS capacity and can reveal ways to decrease rates of consumption through archiving.
Storage is only as valuable as its ability to supply data to business functions. The value of SAN and NAS architectures lies in creating a flexible storage pool that provides a range of storage qualities. During assessment, provisioning, monitoring, planning, and configuration of available storage resources, SRM can coordinate these qualities into unified solutions.
Bill Pierce is the director of software architecture at TeraCloud (www.teracloud.com) in Bellevue, WA.