Isilon on its separate scale-out reality

Why it's different from NetApp ...

Comment Isilon has said it has a NetApp differentiation problem with Sunnyvale's version of scale-out filing conflicting with its own. Meanwhile it is looking appreciatively at deduplication, server flash and running apps on its S-Series.

El Reg was briefed by EMC Isilon's marketing VP Sam Grocott. He said Isilon "recorded around 60 per cent year-on-year third quarter growth before the acquisition and it has accelerated since then. We've seen dramatic growth in enterprises since the acquisition."

Isilon's headcount is up and a brand new office has opened in Seattle. Sujal Patel, Isilon's former CEO, runs EMC's Isilon Storage Division and is a peer to Brian Gallagher, the head of VMAX-associated activities. Both report to Pat Gelsinger.

Isilon and NetApp

He says that business has seen such a dramatic growth in unstructured data that scale-out NAS, previously a feature of high-performance computing (HPC) in the entertainment, seismic data processing and allied worlds, is becoming applicable to big data-type applications in enterprises. Although suppliers such as NetApp have increased the scalability of their file storage offerings they do not scale as well or as easily or with as much performance as Isilon.

Grocott was particularly keen to separate Isilon's scale-out file technology from that in NetApp's Data ONTAP 8.1. Isilon's OneFS filesystem scales up to 15PB, "a self-inflicted limit on its single volume, single file system, introduced at EMC World in May". Theoretically it could go even higher.

He says ONTAP 8.1 is "still a dual-controller architecture in a limited cluster with single name space added to top of FlexVols: multiple volumes." Why does having a single volume matter? What is the disadvantage of having multiple volumes under a single namespace?

"Applications in servers are attached to a single FlexVol. Even though it's a single name space you are limited to a 2-headed controller that's accessing that volume ... You have to run StorNext on top of E-Series to get better scale out for big data ... Not all scale-out is created equal. Other vendor's scale out is heavily compromised: it's a veneer they're presenting."

In other words, applications connected to an Isilon storage resource get far more data bandwidth from the nodes in that resource than they would connected to an ONTAP 8.1 cluster because they can only get bandwidth from a single FlexVol and not from all the volumes in the cluster. That's Isilon's view.

Panasas and DataDirect Networks

Grocott positioned Isilon versus Panasas: "Panasas actually has a true scale-out system, but we are optimised for the enterprise; they're optimised more for Linux/high-performance computing (HPC) workflows. Our enterprise features include everything from snapshot, data protection, auto-tiering with SmartPools, to performance and capacity management tools."

What about DataDirect Networks, which is heading towards enterprise big data applications energetically?

"We do seem them in the vertical HPC markets – entertainment and oil and gas for example. Our differentiation is the integration of things like snapshot and replication; those are plug and play.

"When you install a DDN system you need an IT admin guy with a PhD to understand the complexities of how to tune the system. A lot of time can be needed for the client-side drivers and a lot of admin time can be needed for constant and ongoing tuning to maintain performance.

"Our system doesn't need a PhD-qualified sysadmin; it's plug and play setup, configuration and scale. You can add nodes and scale in 60 seconds. You can't do that with DDN; it's very very complex."

"We don't see BlueArc much – less so since the HDS acquisition. NetApp is our primary competitor." Ditto IBM's SONAS and HP Ibrix.

Isilon hardware running VMs

At a recent EMC event we saw an application in a virtual machine (VM) running on Isilon S-Series X86 hardware, an example of a server app brought to the storage. Sam G said that was a demonstration and that there was no current S-Series capability of running server apps, nor was anything like that on Isilon's public roadmap.

Okay Sam, but we can explore what the demo means. Our understanding is that VMware's ESXi demands sole ownership of the software-hardware interface in a server. That means an S-Series or other Isilon array running server apps would first of all have ESXi controlling the server hardware and all other software running as VMs under ESXi, including OneFS and whatever apps their might be transferred from their natural server home.

We would say that OneFS would need to have guaranteed residency in its X86 engine and not be a swappable VM, for data access speed reasons. We guess that ONeFS could be combined with and so run in ESXi, which would accomplish that, but that route seems very complex. Grocott wouldn't discuss these speculations by the way.

Project Lightning and other EMC ephemera

EMC has its Project Lightning with a VMAX or VNX storage array's FAST (Fully-Automated Storage Tiering) transferring data to a server's PCIe flash memory resource. Is Isilon going to do that?

Isilon doesn't have FAST, having its own tiering software called SmartPools instead. Grocott said: "Re the EMC Lightning idea, the market is going that way ... Lightning is block-focused and we are a file.

"We have announced nothing. Our tiering is specifically built for scale-out files – SmartPools – and we'll continue to invest in that going forward."

El Reg detected a slight hint there that SmartPools could be extended server-wards in the future and manage a server's PCIe flash resource.

What about the idea of running (X86) OneFS code as a storage personality on VNX hardware? Grocott didn't actually flinch at this heresy, and said: "Our node architecture is very different from VNX."

We take it there is no intention to port OneFS to be a personality on VNX or VMX base and use the host controller's storage enclosures.

How about deduplication? He said: "We absolutely believe deduplication is an important factor going forward in the scale-out market. It needs to be designed in and not just bolted on. We have not announced a release at this point but we think its important for the big data, scale-out NAS world going forward.

"To do this correctly it has to be designed in from day one to avoid making tremendous compromises to the applications."

Okay, Isilon has deduplication coming and it doesn't appear to be Permabit's Albireo but rather an in-house Isilon development.

How about the idea of an Isilon storage presence in EMC's Vblocks, an Iblock for example? Grocott said: "Isilon is addressing exclusively the storage tier in big data, scale out file applications." In our interpretation of his view, the Vblock approach is not relevant.

Isilon is booming under EMC's custodianship. El Reg sees it is as semi-autonomous operation, able to develop its own technology in an EMC-compatible way, and take advantage of EMC IP and ideas. And also, of course, of EMC's enormous sales channel. Glory days are heading Isilon's way and it is itching to have them come as fast as possible. ®