Like most conventional wisdom, it's based on the past and not the present. In other words, the conventional wisdom is wrong.
New technologies and architectures are providing performance that is equal to or better than the best SAN storage arrays. For the best per NAS system performance there's BlueArc's Titan 2. It can demonstrate as much as 300,000 NFS IOPS or approximately 98,000 SpecSFS IOPS per system (600,000 NFS IOPS or 195,000 SpecSFS IOPS per clustered pair) and throughput as high as 1.25 GBps per system (2.5 GBps per clustered pair).
NAS scalability has been addressed with three different technologies that scale both performance and capacity. These technologies include scale out software-based clustered file systems, scale out hardware and software-based clustered file systems, and global name space aggregation (GNS).
Scale out software-based clustered file systems (alphabetically listed):
- Polyserve (HP sells this as their Enterprise File Services Clustered Gateway)
The clustered file system methodology varies by vendor. Exanet and IBRIX both provide the NAS services, as well as the clustered file system. Polyserve provides the clustered file system for standard Microsoft Windows Storage Server 2003 or Linux file servers. Hardware is usually industry standard x86 servers. The software provides near-linear scalability for performance and capacity. Cluster limits are a combination of diminishing marginal returns (when the additional clustered node only marginal increases the scalability because of increased clustered overhead) and testing.
Scale out hardware and software-based clustered file systems (alphabetically listed):
- BlueArc (Titan 2 N+1 clustering)
- Cloverleaf (iSN N+1 clustering available now)
- Isilon (N=1 clustering available now)
- NetApp (GX series to be available later this year)
The number of units that can be clustered varies by vendor, testing and software clustering overhead, which activates the law of diminishing marginal returns. BlueArc is currently two nodes and will be going to 32. NetApp is currently two nodes and going to eight. Cloverleaf is currently two-to-12 nodes and going to 1,024. Isilon is at 42 nodes and increasing continuously. Hardware is proprietary in the case of BlueArc, Isilon and NetApp, and standard AMD x86 and Solaris architecture in the case of Cloverleaf.
Global name space aggregation (alphabetically listed):
GNS is essential an abstraction layer above the NAS systems it aggregates. The scalability varies by vendor (billions of files for Acopia, tens of millions of files for the others). GNS provides a single image for mounting load balancing and data protection. It allows nondisruptive data migration between NAS systems as well.
NetApp set the standard for storage data protection simplicity and functionality with their SNAP brand of services. Most NAS vendors and products have responded in kind. In general these data protection services are easier to use, provide more options and have better RPO/RTO granularity than Enterprise SAN storage.
In my view, NAS has definitely evolved. It's still the simplest storage to implement, operate and manage. Now it has the performance, scalability and data protection equal to or better than Enterprise SAN storage as well. Next time you need to purchase storage, it is could be well worth your time to take a look. The money, aggravation and headaches you save may be your own.
About the author: Marc Staimer is president and CDS of Dragon Slayer Consulting in Beaverton, Oregon. He is widely known as one of the leading storage market analysts in the network storage and storage management industries. His consulting practice of six plus years provides consulting to the end-user and vendor communities.