Isilon adds faster clustered file storage
Deduping nearline storage too
Isilon has souped up its mainstream clustered NAS product and introduced a transactional IO product with SAS disks plus a lower-cost archive model with optional Ocarina deduping.
Isilon has made its name with scale-out network-attached storage (NAS) X-Series products which scale to 2.3PB with 96 nodes clustered across InfiniBand, all inside one global OneFS filesystem volume and namespace. The products use a 1Gbit/s Ethernet host interface and 7,200rpm 1TB SATA drives, and have proved popular in digital media file processing applications such as special effects rendering in movies and other high file-access rate environments such as the pharmaceutical, life sciences and Web 2.0.
In early 2008 Isilon swapped out its CEO and CFOs and restated its accounts. Co-founder Sujal Patel took over the CEO reins and has been clearing up the financial situation since then and recruiting new senior executives to run product engineering and operations and also the sales channel - by recruiting Leonard Iventosch, NetApp's ex-channel sales boss.
Fortunately for Isilon, whose product strategy is predicated on file data storage needs overtaking and then dominating block storage needs, its main competitor NetApp has had its own difficulties developing a scale-out NAS clustering product. NetApp's ONTAP 8, which is hoped to deliver that functionality, won't be released until later this year.
This has given Isilon time to strengthen its core offering and introduce two complementary product flavours to encourage Isilon customers to spend more money with the company.
The new mainstream X-Series box is the IQ 36000 which uses a punchier controller, featuring dual quad-core 2.33GHz Xeons in a 4U enclosure. It offers up to 36 1TB SATA drives per node and four 1Gig E. The minimum cluster size is 7 nodes and the maximum cluster of 96 nodes can support 3.45PB of capacity. It has 10GigE support when combined with an IQ Accelerator node and offers more than 400MB/sec single stream I/O performance.
The 5400 S-Series node takes advantage of 15,000rpm SAS drives to offer a faster response per node to file access requests for more transaction-focussed workloads. Each 2U node can have 12 450GB SAS drives, that's 5.4TB/node, and use the same controller as the IQ 36000 to support 12,000 IOPS/node. A 5400 cluster can scale from 16TB to more than 500TB, have 1.5TB of globally coherent cache, and support more than 1 million IOPS. Isilon reckons the 5400 S has 30 per cent better price/performance than NetApp, its main competitor, at its list price of $2/GB.
By taking performance-accelerating features out of the IQ 36000, Isilon has been able to reduce its cost and produce an IQ 36NL model variant suited, it says, for nearline or secondary storage applications. The controller has a single quad-core Xeon with only a single gigE host interface port and less memory at 4GB of SDRAM versus the 36000 X-Seres' 8GB. Its capacity per node is the same but the cost per node much less.
Isilon has also made Ocarina deduplication available. This is digital media file-focussed deduplication with special mathamatical techniques used to compress the supposedly uncompressable, JPEGs and the like. It marks Isilon out from what is becoming a run of the mill deduplication market which it would have been solidly planted in if it had licensed software dedupe technology from, say, FalconStor or Quantum, or OEM'd a hardware dedupe box from Sepaton. These are all good dedupe offerings but becoming middle of the road in a sense now, which Ocarina is not. Ocarina's digital media focus also matches a certain segment of Isilon's customer base.
The 36NL also provides asynchronous replication to a remote site for disaster recvovery purposes.
Isilon’s S-Series, NL-Series and X-Series products are all immediately available. The 5400S is listed at $49,999 per node, the 36000X at $137,000 per node, and the 36NL at $72,000 per node.
Isilon has lots of future headroom. The current quad-core Xeons could move to 8-core Nehelem processors. The 20Gbit/s InfiniBand link can progress to 40Gbit/s. The 1TB SATA drives can move up to 2TB ones. Solid state drives can be introduced. We might envisage Isilon clusters approaching 200 nodes and 8PB in size as a consequence over the next 2-5 years.
Isilon's products support far bigger volume sizes than mainstream NAS products and this, Isilion says, makes the volumes far easier to manage, leading to much lower costs as well as enhanced performance. The company has managed to show stable revenue and customer growth despite management and funancial hiccups. In 2007 678 customers generated $89m of sales while in 2008 922 customers generated $114.4m. Gartner has given it scale-out NAS market quadrant leadership which is a nice boost. Who knows, if it can continue to grow sales in the recession, it might even make a profit soon. ®
$2/GB my arse.
$2/GB my arse. Thats a mulched raw figure at best. I have seen a recent list price quote from these guys for way over $200k+ to protect as little as 30TB.
As for the de-dup stuff, is that algorithm guaranteed by legislative courts to provide authenticity of the data when its rebuilt? No. Clever it may be, further ways to tie you to a vendor it is. Buying it, we will not be.
"This is digital media file-focussed deduplication with special mathamatical techniques used to compress the supposedly uncompressable, JPEGs and the like."
Of course JPEGs can be recompressed. In fact virtually any media format file which uses lossy compression, like JPEGs, MP3s, MPEGs and so on can be reduced in size at the costs of quality. No special mathematical (or even mathamatical) techniques required. Alternatively, you can perform a file conversion to a more compact format and regenerate the original data type (rather naughtily, some software used to do this with non lossy files such as TIFFS - they would convert to JPEG before storing and regenerate the TIFF, thereby losing detail. That was important on medical archival systems for things like X-Rays where detail can literally be life-and-death).
However, what is utterly impossible is to do a "lossy" compression without being aware of the data type is. Do that and you have an unreadable file.