Many NAS faces: Which one is right for you?

NAS offers many of the same uses as SAN today

1 2 Page 2
Page 2 of 2
  1. Clients and servers access data from "virtual servers

  2. Multiple virtual servers can reside on a single NAS gateway

  3. Multiple NAS gateways can exist within a single cluster

  4. Balance workloads across NAS gateways in the cluster by moving virtual servers

  5. To scale performance, NAS gateways can be added to the cluster

  6. Virtual servers enable isolation and autonomy of applications and workloads

Capacity scalability: Capacity is easily scaled as well. NAS gateways are physically connected to a pool of storage so capacity can be added at any time by adding more disks to the storage pool.

Parallel-access NAS characteristics:

  • File system spans all nodes

  • Data accessed from all nodes simultaneously

  • No snapshot capabilities

  • No virtual servers

  • No automated storage provisioning

  • May require special host or client software

Scalable sequential throughput
Because data is accessed from all nodes in parallel, sequential throughput (that is, the bandwidth available to move a single file) is scalable. While this scalability can be a benefit for data streaming environments, in enterprise NAS applications it actually becomes a weakness.

The reason is this: In a parallel NAS access architecture, all data accesses affect all nodes. Every I/O request generates workload for every node, so random throughput becomes limited by the performance of a single node and does not scale as nodes are added. 

A variant of parallel access NAS can provide scalable random throughput. Known as clustered file systems, this approach is similar to parallel-access NAS, since a file system can span a number of nodes. Hence, the cluster can be configured to deliver high streaming throughput or high random throughput. For a given file system, the administrator selects either sequential or random throughput as the priority.

Advanced management features are usually lacking in the clustered file system implementation. Because these products are simply file systems (rather than complete appliances), management capabilities will be no different than what was found on the underlying server. If multiprotocol support, variable I/O workload, snapshots, open storage and ease of use are important, clustered file systems are not the optimal choice.

Feature set reflects the environment
In general, the feature sets of the clustered and parallel access technologies are slimmed down to focus on the performance needs of HPC environments. Since HPC applications do not require the complete feature set of traditional NAS such as data snapshot, robust multiprotocol support, and automated volume management, these capabilities may not be present.

Which is right for you?
As NAS architectures continue to evolve, there are various considerations for each approach, as reflected in Table 2. Legacy NAS excels in simplicity, yet lacks scalability, while clustered NAS gateways and clustered or parallel-access NAS implementations overcome scalability limitations in different ways. Next-generation architectures provide IT managers with an opportunity to leverage NAS for a new class of applications, and understanding the differences in approaches is critical to deploying the correct option for each specific storage environment.

Table 2: NAS Architecture options comparison

Clustered NAS Gateways Legacy NAS Clustered and Parallel-Access NAS
Benefits Scalable random throughput with on-demand capacity and performance. Virtual servers permit load balancing and rapid deployment for workload changes. For scaling, open external disk is supported. Data protection includes snapshots. Easy to use, self-contained appliance. Some implementations have peer or active node for file system and workload fail-over. Sometimes referred to as clustered nodes. Includes data snapshots. Scalable sequential file system throughput. Utilize Windows or Linux running on low-cost or purpose-built hardware (parallel file system) to support HPC and other technical environments.
Caveats Single file system with read/write throughput limited by single node. (May support mount of file systems as read-only from multiple nodes.) File system performance limited by single node. Limited capacity and performance with in the appliance device. Optimized for sequential vs. random performance. Complex administration (for parallel file system). Snapshots capability not always present. Solution geared toward technical environments. Host software may be required.
Best uses Windows file server consolidation, home directory, web content delivery, commercial random I/O applications, engineering/software development, file serving. Basic file serving and data sharing, Web serving, small databases, departmental and workgroup storage. Video postproduction, computational fluid dynamics, energy and mineral exploration, scientific.
Examples ONStor NetApp, EMC, Blue Arc Isilon, Polyserve, Ibrix

Items to consider when selecting an NAS approach include:

  1. Application workload: Is the workload generated by multiple clients and servers?

  2. Organizational issues: Will groups want autonomy and control their own servers?

  3. Multiprotocol: Will multiple platforms (Windows and others) be supported?

  4. Snapshots: Is there a need to roll back in time to recover files?

  5. Multivendor storage: Can existing SAN or other disks be redeployed to support NAS?

  6. Performance: Will files be accessed randomly or sequentially with large or small I/Os?

  7. Ease of use: What are your requirements for ease of use, and what skill sets can you leverage?

  8. Growth: What are your growth plans?

Greg Schulz is founder and senior analyst of the StorageIO group  and author of the book Resilient Storage Networks (Elsevier).

Copyright © 2006 IDG Communications, Inc.

1 2 Page 2
Page 2 of 2
7 inconvenient truths about the hybrid work trend
Shop Tech Products at Amazon