What's next for NetApp hardware?
Storage and the PCIe switch
NetApp has confirmed it is working on new hardware platforms.
Rich Clifton, a senior VP at NetApp, provided a glimpse of NetApp's hardware direction this week at VMworld, as The Register sought to determine what storage OEMS would want with a PCIe switch.
Clifton said there would be phased ONTAP G announcements, suggesting that functionality could be delivered in stages. ONTAP 8 delivers clustering to the NetApp world and is the merger of the existing 7G and GX ONTAP variants, ONTAP being the operating system for the bulk of NetApp's storage products.
The senior VP said he knew nothing about PCIe use by NetApp but did say that any interconnect is chosen first on three parameters: bandwidth, latency, and cost. He said that current FAS 6000 clustered pair products use an InfiniBand point-to-point link between the two storage processors. In fact, NetApp is a significant shipper of InfiniBand links because of this.
Virtensys, another operation we spoke to at VMWorld, makes a VMX-5000 switch that connects to X86/PCIe servers by external PCIe cables. This lets them share network adapters (such as FC, IB, and Ethernet) as well as storage mounted in the switch. The attached servers each see virtual adapters and virtual direct-attached storage and think it's all local.
Marek Piekarski, chief technical officer for VirtenSys, says it's talking to potential server and storage OEMs. It's understood one server OEM has provisionally signed up. After NetApp was introduced into the conversation, he said that we could think of a scheme of storage processors in a matrix connected by PCIe. He didn't actually say they were talking to NetApp though, so what follows is supposition.
There's no need for a VirtensSys switch product in this environment unless the storage processors connect to the switch and share the I/O cards in it and/or use the switch to simplify a mess of point-to-point links, meaning more than two storage processors, Perhaps much more than two, heading towards five or more.
Piekarski said they were working on a second generation of the switch technology to add inter-processor communications capability so it would have InfiniBand-like low latency.
This nets out to a possible high-end NetApp FAS-bigger-than-6000 box which has a cluster of five or more storage processors linked by a low-latency PCIe cloud. All would talk to the outside world via a VirtenSys gen 2 switch providing the IPC capability, and there would be shared adapter use so that the box could talk FC, FCoE, iSCSI, or NAS to connected servers.
Each storage processor would, logically, have its own storage enclosures with, potentially, fast disk, slow disk, and perhaps solid state drives in them. Some storage processors might, though, be doing storage management things and not actually manage their own local storage.
How about that for an ONTAP 8 box?
If NetApp were to vary the HW specs of the storage processors, it has an obvious way to have faster/slower versions.
Getting back to VMworld, such a new box would have VMware vStorage API interface code added to it, so that it could play well in VMware's shiny new vSphere data centre environment.
The net net of this mental Lego construction exercise is that a FAS 7000 or 8000 with clustered storage processors could be announced in the 2009/2010 timeframe. It would be modular storage but go way beyond the classic dual-controller arrays we have gotten used to and take NetApp up into the Symmetrix area and/or into the Data Direct/Isilon product area of very fast clustered storage.
Take this with a pinch of salt though. VirtenSys could be talking to other storage OEMs instead of or as well as NetApp. ®
Sponsored: Benefits from the lessons learned in HPC