Some time ago, November, 2003 in fact, Network Appliance bought Spinnaker Networks. Since then the world has been waiting for NetApp to use Spinnaker technology to group distributed NetApp storage products together. At the time Dave Hitz said buying Spinnaker was all about grid computing for business. Spinnaker had developed technology for distributed systems architecture, clustering and virtualisation.
But nothing happened. Spinnaker's product was the Spin Server. Perhaps that had infected NetApp marketing and all we were hearing was spin, spin, spin.
Not so. NetApp has had a real job of work integrating the Spinnaker SpinOS technology into Data ONTAP, the operating system that runs in all its products. Last November, NetApp CEO Dan Warmenhoven talked about: "the complexity of integrating millions of lines of code." NetApp's FlexVol virtualisation technology has to be integrated with Spinnaker's Virtual File Systems (VFS - see below). Indeed, in fact we could think of the FlexVol and VFS concepts being combined into a kind of virtual FlexVol file system or, alternatively, FlexVols existing underneath a VFS layer.
Data ONTAP 7G offered multi-vendor storage array virtualisation to NetApps gFiler products. Now Data ONTAP GX has been installed for what is effectively late-stage beta testing at several customer sites. This version of ONTAP uses the Spinnaker distributed systems technology.
It's a high-performance O/S and beta customers include companies in high performance technical computing markets (HPTC) such as energy, ECAD, and entertainment.
NetApp also demonstrated the new software at its annual Analyst Day event. It ran in an eight-node NetApp FAS3050 configuration and, NetApp said, showed that multiple NetApp storage systems can be managed as a single entity under a global namespace. (Interestingly NetApp used the mid-tier FAS3050 and not the high end FAS980, yet Data ONTAP GX is supposed to be a high-performance O/S.)
It means that all members of an application server cluster can access data stored across all of the FAS systems by using a single access point. You don't have to map individual application servers to storage systems.
It also means that you increase performance and capacity by adding more FAS storage boxes - scaling out. Data volumes can span multiple storage nodes and so aggregate performance can increase.
There is virtualisation involved too as Data ONTAP GX has the ability to dynamically add storage resources and transparently redistribute data without any disruption to client systems.
NetApp says the result is a storage system that combines the advantages of management simplicity with scalable performance for compute-intensive applications.
However there may have been a problem with resilience; failover to other boxes when a FAS storage box goes down being a tricky technical problem to solve. The Salk Institute abandoned a SpinServer installation because of failover problems.
Where does the 'GX' come from? In October, 2004, NetApp supplied SpinServers to GX Technology, a Texas-based seismic data processing business. Then SpinServer was 'a flexible, high-performance grid storage system that can scale to encompass as many as 512 cluster nodes and a total of 11PB (11,000TB) of storage in a single installation.'
The NetApp case study goes on to say: "network clients ... view all storage on any of GXTs 16 nodes as if it were a single large file system and access it all with a single mount. Virtual file systems (VFSs) stored on individual nodes can be transparently migrated between servers to balance workload and optimize performance without disrupting ongoing work."
Also: "The great flexibility and ease of use of the system have allowed GXT to redeploy one of the two administrators formerly responsible for the Sun servers. A single individual manages the 250TB SpinServer installation plus performs other duties. Because of cost savings due to decreased management costs, decreased server acquisition costs, and increased performance, GXT estimates that the SpinServer deployment has resulted in at least a tenfold gain in overall price/performance over that of the previous environment."
GX may simply reflect the experience NetApp gained through working with GXT.
Net App says in its GXT case study: "A SpinServer cluster is a group of individual SpinServer nodes connected through a dedicated cluster interconnect (typically Gigabit Ethernet). Each SpinServer node controls one or more storage pools. Every storage pool, in turn, contains one or more virtual file systems (VFSs). VFSs are directory trees, each consisting of a root directory and a set of nested subdirectories and associated files."
"A SpinServer cluster exports a single global file system by arranging the complete set of VFSs from all SpinServer nodes into a single tree. One VFS is identified as the root VFS, which defines the root directory of the global file system. All other VFSs are joined to the global file system in a designated manner. A network client sees a single large file system and can access any part of the global namespace without having to mount the specific SpinServer node that stores the desired data. The aggregate bandwidth available from the single global namespace scales linearly with the number of systems in the cluster, making SpinServer ideal for applications where high bandwidth is critical."
It's a very fancy network file system (NFS). The sysadmin can move VFSs around the cluster to balance workloads. He gets the data copied over and then remaps pointers to the new VFS instance. The old one gets released.
When more storage bandwidth is required, another SpinServer node can be added to the cluster without bringing the cluster down, and workloads can be transparently migrated to that system, which becomes part of the global namespace. If an individual system requires maintenance, its VFSs can be transferred to the other cluster members (assuming they have enough free storage space) so the system can be serviced without disrupting operations.
The theory sounds good
Data ONTAP GX theory sounds good. But it is already two and a half years since the Spinnaker acquisition, a long time. NetApp is betting that grid storage will become a standard. It can't just be a high-performance computing niche technology in NetApp's view; there's no point in re-architecting your entire product line's storage O/S unless there is going to be wide-spread takeup.
For NetApp storage virtualisation and management resides in its products, not in fabric switches. That's a concept collision point with EMC and its fabric-based InVista approach, also with IBM and its SVC. (Interestingly IBM has a reselling agreement with NetApp. Expect some kind of event here concerning IBM's SVC and NetApp's Data ONTAP GX.)
The large-scale storage supplier world is splitting into two camps here: the fabric intelligence duo of EMC and IBM and the storage controller-based virtualisation trio of HDS, NetApp and Sun. HP, since it OEM's the HDS TagmaStore is aligned with this trio as well although it is by no means evangelical about it and seems to be, at heart, agnostic on the issue.
We can envisage both Fibre Channel disk and SATA Disk boxes in a GX storage grid with the SATA boxes used for archiving, disk to disk backup (think Kazeon here) and fixed content storage. We can also envisage multi-vendor arrays in a NetApp storage grid: NetApp boxes; HDS arrays; EMC ones; and IBM and Sun ones too.
When will we see product?
At a February, 2006, investment call, Warmenhoven said: "We will release the first version of ONTAP GX, as on option to FAS 3000 in a very near future." Also he said: "Youll see it in the March 14 analyst day, and well be right on the cusp of, of making that a generally available release, or available to any customer who wants, I should say. GA is a different term, but FCS release, right at the end of this fiscal quarter."
That means April 1st!
It will be delivered thus: "The customer who chooses a (FAS) 3000 or one of the new high end systems (Excelsior - above the FAS980) will have their choice as to which operating system they want installed when its shipped, and customers who have current 3000 systems have the opportunity to upgrade those if they so choose. That upgrade process requires that they have FlexVols enabled on their current ONTAP environment, and from that point it, should be a fairly straight forward process."
Both GX and 7G versions of Data ONTAP will be available. As GX comes to market as a subset of the total ONTAP capabilities, NetApp will be adding additional features to it over time, possibly a couple years. We might expect 7G and GX to combine two or three years out into a single, grid-enabled, NetApp storage O/S.
Find your next job with techworld jobs