Just as there is a debate in the storage area network (SAN) world about the best point at which to virtualise SAN storage, a similar debate is ongoing about network-attached storage (NAS) virtualisation. In the SAN case the debated locations are either in the fabric or in a clever controller front-ending the storage arrays. In the NAS case the three locations are the accessing servers (i.e. PolyServe), a clever switch (i.e. Acopia) or a clever set of NAS boxes (i.e. NetApp).
We looked at Acopia's offering here. The two alternative approaches involve having accessing servers to share a clustered file system - the PolyServe way - or having NAS boxes clustered together and sharing a global namespace. NetApp has just announced its clustering NAS product - Data ONTAP GX. We'll compare and contrast the PolyServe and NetApp approaches in this article.
With PolyServe's File Serving Utility (FSU) product the backend storage is block data on SAN disks. But it is not seen as a SAN by the accessing servers. They see it through the filter of a file system that runs on each one of them and which clusters them together in a single file system. The same file system is mounted simultaneously on many servers. You can add more servers (scale out) and there is a failover facility if any server should fail.
The clustered servers access the disk storage via either NFS or CIFS - the cluster has to be all Windows or all Linux. No other version of Unix is supported by the way. Also any direct-attached storage belonging to each server is not included in the clustered file system.
The disks are accessed over a Fibre Channel SAN fabric. PolyServe has added iSCSI support in the past couple of months and Chris Gomersall, PolyServe GM for EMEA, said: "We work up with EqualLogic as an iSCSI target." He has good things to say about EqualLogic.
To call them clustered servers may be stretching a point; there isn't any high-speed cluster interconnect. Instead ordinary Ethernet is used and distributed lock management and other cluster traffic runs across it.
In effect all the servers clustered with FSU form a single logical NAS head. It seems unpromising; take a bunch of servers running applications and have them a distributed NAS head/clustered file system too. But in fact it is a very sensible thing to do. There is generally lots of processing head room in servers and bring distributed server/NAS islands together in this way.
You now have just one logical NAS resource to manage in one place. It is one resource to back up and to protect. With a distributed server/NAS set up you have a multitude of NAS systems to back up and protect - much more demanding of management time and effort.
Individual NAS islands can run into scaling difficulties - requiring expensive fork-lift upgrades. With PolyServe's FSU you just add more disks. It's simple. Also you are not restricted to a particular NAS supplier's drives.
Gomersall described a second use model for PolyServe; server consolidation with MS SQL Server. This lovely product can't run well under VMware; it just makes too heavy demands in terms of I/O and processing needs. It's a well-known problem apparently. If SQL Server instances run in servers in an FSU cluster then they run at a higher utilisation rate and you can put multiple SQL Server instances on a smaller number of servers.
Big UK banks are interested in this scenario, according to Chris Gomersall. One consolidate 400 SQL Sever servers onto 50 servers with PolyServe, which paid for itself within one year.
Fidelity Investments is one of PolyServe's UK customers. Another is HSBC. In the USA Amerada Hess uses PolyServe's FSU. Vic Forsyth, Manager, Exploration and Technical Systems for Amerada Hess Corporation, is quoted on PolyServe's website: "To match the performance and capacity of the 60-terabyte, fully redundant PolyServe NAS Cluster, we would have had to purchase a traditional NAS solution costing seven times the price of the PolyServe solution and incurred a significant administration headache."
PolyServe helps businesses with distributed NAS islands and a wish to consolidate them and/or servers running SQL Server. Through the FSU they gain NAS scale out headroom, server scale out headroom, simpler and more effective NAS resource management, and SQL Server efficiencies.
What need is there for NetApp's clustered NAS system then? Where does Data ONTAP GX fit?
NetApp ONTAP GX
It doesn't fit into either of the PolyServe use models. That's because it is aimed at the high-performance computing (HPC) market where applications are both processor cycle-sensitive and I/O-bound. You don't bog down these servers further by having them run NAS head functionality. What you do is provide them with a very high-performance clustered NAS product, one that can scale out in capacity and performance terms.
That's what NetApp has done with ONTAP GX. It has developed the new product specifically for HPC customers and will move general business-focussed NAS functionality from ONTAP 7G to GX 'over time.' For now NetApp's general business customers are recommended to use 7G; the HPC ones are pointed to GX.
GX customers have their servers 'see' a single NAS resource presented to them by clustered GX 'nodes' which provide scaling ability, high-availability and data protection - RAID 6 double protection and snapshot functionality. Work can be load-balanced - it can be with PolyServe but it is a manual operation.
Where is NetApp going with 7G? Is it going to stay in an HPC ghetto?
Indeed not. Over time all NetApp NAS products will be capable of being clustered. That means that the generality of NetApp business customers will be possible candidates for a clustered NAS system. At present, if the suffer from distributed NAS island sprawl the PolyServe clustering FSU provides an alternative to NetApp NAS upgrades, but at the cost of server CPU cycles. However PolyServe can go further, arguably, with NAS consolidation for business users, than NetApp can, at present.
What happens when the V-filers run GX is going to be pretty interesting, because then V-filers can be clustered. Heterogeneous arrays could then be used too.
NetApp's approach is being targeted at high-end NAS needs, ones in the high-performance computing space, It is intending to move missing business functionality from ONTAP 7G into GX at a later date. Until then business wanting to virtualise NAS resources must look at other vendors.
PolyServe's approach has won many adherents, not least in the UK which now represents 25 to 30 percent of PolyServe's global business. HSBC is one customer and the storage consolidation and management simplification that PolyServe brings is attractive to it.
However, with an accessing server-based NAS virtualisation and clustering approach you inevitably have accessing servers doing two things: running the applications they were bought for; and virtualising and managing the NAS resources. If there is the server processor headroom and network bandwidth to do this then the approach is good. It's certainly an advance of distributed NAS islands.
But, much as SAN virtualisation has moved away from servers, so too may NAS virtualisation. Doing it in a set of clustered NAS boxes has a strong performance edge, as both NetApp and Isilon exemplify. If NAS performance is what counts and your accessing servers don't have the spare capacity needed then look instead at virtualising NAS resources in the NAS layer itself.