Isilon makes grab at NAS performance crown
SPEC claims drowned out by bickering
Isilon's claimed SPEC benchmark score has shown just how good Exanet systems are - says Isilon.
Isilon, which provides scale-out NAS clusters, has published the results of a Standard Performance Evaluation Corporation (SPEC) sfs2008 NFS benchmark for its IQ 5400S product. A 10-node 5400S cluster delivered 46,635 IOPS, with a corresponding overall response time of 1.91msec. The company says it is the first scale-out NAS supplier to publish a SPEC sfs2008 benchmark and has set a performance bar.
But both claims are being hotly disputed by a rival.
Exanet also provides clustered NAS systems, scaling to 1PB by adding nodes, and has published SPEC sfs2008 NFS benchmark results. In August 2008 its ExaStore 2008 Clustered NAS achieved 119,550 operations per second with an overall response time of 2.07 milliseconds, a world record. The Isilon system has not exceeded that.
Exanet benchmarked an 8-node ExaStore Clustered NAS System, based on ExaStore 2008 software, EX1500 NAS Servers, and DX12 storage arrays. It had 224GB of memory, and 592 disk drives with 64.5TB of capacity. The EX1500 servers, like the Isilon 5400S, use dual quad-core Xeon processors.
Isilon's 10-node cluster had 165GB of memory and 120 disk drives with a 48TB capacity. What this comes down to is that the 10-node Isilon cluster, with five times fewer drives and a third less capacity, delivered 39 per cent of the performance of an 8-node Exanet cluster with many more disks but greater capacity. It looks like spindles count, much more so than compute horsepower.
Sam Grocott, Isilon's senior director of product management, said: “(Isilon is) the first and only scale-out NAS system to publish the SPEC sfs2008 benchmark."
With Exanet having published its sfs2008 results last year and comfortably exceeding Isilon's sfs2008 performance this claim, and the performance bar one, both seem to be incorrect.
A person familiar with the situation said: "Absolutely not. Isilon is definitely not the first."
Nir Peleg, one of Exanet's founders, checked the Exanet and Isilon SPEC sfs2008 benchmark reports and confirmed the numbers above: "It seems to be all correct."
Asked if Isilon was wrong to claim to be the first scale-out NAS vendor to publish sfs2008 benchmark numbers and also wrong to claim to set a performance bar, Peleg said: "Right." He added: "We didn't tune our systems and we think we could achieve the same numbers with fewer disk drives."
Isilon is standing by its claims.
Sam Grocott said that Isilon doesn't consider Exanet has a true scale-out NAS architecture. As defined by Gartner and others it means scaling out performance and capacity in a single file system and a single volume. He said Exanet's architecture involves multiple volumes with eight or twelve clustered server heads and hundreds or thousands of disk drives stacked behind them. There are multiple volumes and it's a traditional clustered NAS architecture, not a true scale-out NAS concept.
He said an Isilon cluster could scale out to 144 nodes and about 650,000 IOPS, more than five times higher than the Exanet number. ®
Exanet - single FS per cluster
As an Exanet customer I can say their systems have 1 file system per cluster. You can have multiple volumes in that file system but there's no requirement for such. The downside is there is only 1 file system per cluster, no storage tiering - yet.
As the article implies most of the performance comes from the spindles, most NAS systems are "fast enough".
Exanet had another customer that did informal testing last year showing 3PAR back end storage providing double the per-spindle (NFS) IOPS as their own tier 2 storage. As a customer of both companies I've been hounding them off and on to do a joint SFS benchmark to show what the combined system can do(since they pimp each other's gear quite heavily around the country). I'm confident that Exanet can achieve the same(or better) high end IOPS numbers with roughly 300 drives on a 3PAR vs 500+ on the tier 2 stuff. Time will tell though. The better architecture, caching, and controllers on the 3PAR just destroys that tier 2 crap. (Exanet will tell you this themselves as well).
I certainly like the flexibility of being able to share the same spindles on the NAS with Exanet as I do on the SAN with 3PAR, having volumes striped across every spindle instead of having volumes composed of physical spindles and having to dedicate them to a particular server/cluster. Which is one(among tons) of reasons why I liked Exanet(hardware agnostic for the most part, fast too) and 3PAR(fastest block storage in the west, and east, and north, and south).
Too many folks take back end storage for granted it seems, most people think it's all about the head unit. The back end is at least as important if not more so than the front end. BlueArc finally admitted that to us last year after months of talks with them(used to be a BlueArc customer), they said that their tier 2 stuff(same gear many companies re-sell or OEM) just isn't suitable for performance intensive and really high availability. They go as far as completely disabling the write cache on their tier 2 storage systems because it's _that_ inefficient.
Cooking on low energy gas?
46635 OPS over 10 clustered nodes? I work that out as 4663.5 OPS a node.
Looking at other configurations from other vendors that seems a bit slow... especially with all that disk and RAM in that configuration.
They've set a benchmark, but they've set a very low one.
Anyone seriously bothered about their NAS performance will have bought BlueArc.