Because Compellent's SAN already breaks the data up at the block level in order to spread it evenly among disk spindles, the system holds metadata in cache that tells it how to "reassemble" the data when an application calls for it. From there, according to Bruce Kornfeld, vice president of marketing for Compellent, spreading the data blocks across different classes of disk was a short leap to make.
As such, the new feature for Compellent requires no additional software -- users who wish to use it pay a fee to "unlock" the feature. "We ship all our systems with all features, it's just a matter of which ones the user wants to unlock," Kornfeld said.
Finally, since the feature is part of the SAN itself, it supports data from any operating system (OS) and application, Kornfeld said. "Typically with third-party data classification software you need a different version for each OS environment."
Compellent's marketing collateral estimates that users will end up with just 20% of their data on Tier-1 storage and estimate a reduction in a typical user's spending on disk of up to 74%.
Even if these claims aren't borne out by wider use, many users might welcome the feature just because of its simplicity. Over the last six months, there has been no shortage of users willing to talk to SearchStorage.com about their struggles to manage tiered storage and information lifecycle management (ILM) (See sidebar.)
"Tiered storage is not something I could imagine managing manually," said Bill Snow, IT director for Moss & Associates, a Florida construction company that has grown in the past two years from a five person firm with $25 million in annual revenue to a 300 employee operation with $750 million pouring in each year. Its storage needs have increased accordingly, from a Microsoft Small Business Server with one drive to a 5.3 terabyte (TB) usable capacity Compellent Storage Center SAN, installed a month ago with 500 GB used. In just one month, as new jobs have continued to open up for the company, the data has doubled to just over 1 TB.
The company now boasts six people in IT, as opposed to just one a year ago, but according to Snow, "We're shipping out 10 computers a week to new employees or for new projects. That might not be a lot if you're a Fortune 100 company, but we are so busy, constantly."
Currently, Snow said, the company has 260 GB of data on 10,000 rpm Fibre Channel drives in a RAID-10 configuration; 78 GB on Fibre Channel in RAID-5, which he counts as Tier-1a; and 640 GB on RAID-5 SATA, which he calls Tier-3. Right now, Snow said, he anticipates that Moss & Associates will only have to invest in additional SATA disk for the foreseeable future.
He couldn't give a specific estimate of what the system had already saved him in disk costs, but if he maintains his current distribution of Fibre Channel and SATA, and eventually fills the SAN to capacity, he could save up to $33,000 versus the cost of a single-tier SAN.
Snow also said that he appreciated the system's granularity. "What sold me was that it was done at the block level," he said. "That's great for an application like Exchange where if you're migrating things at the file level, the Exchange Store is one big file -- it wouldn't work."
A reporting and capacity planning tool shows which volumes are taking up the most space in order to anticipate what disk to buy and how much.
"I'm using it on almost every volume on my SAN," said Jim Hanrahan, computer technician and trainer in the office of information services for the Kenosha Unified school district in Illinois.
Hanrahan's 6 TB capacity Compellent SAN has 10,000 rpm SCSI disks in a RAID-10 configuration for Tier-1 and the rest is on SATA for Tier-3. Hanrahan was unsure of the percentage of data on each tier.
"The numbers are skewed because at the end of the year, teaching staff are asked to cleanup their emails," Hanrahan said. "This has had the effect of moving quite a bit of data back up into Tier-1 storage."
What about performance? Kornfeld pointed out that there would be no more performance head associated with "assembling" the file since the blocks are broken up anyway.
"It might take just a little bit more time for a file to be read [that has blocks on lower tiers of storage]," he said. "But we're talking six milliseconds instead of three. Some customers are also reporting a boost in performance on Tier-1 drives because they're not as clogged."
And what if something goes wrong? With data broken up into tiny pieces, spread across not only drive spindles but classes and types of disk, is there more of a risk of data loss? It might give some users pause, but so far beta testers and analysts say the risk is probably not a lot greater than for any storage subsystem.
"I'm very confident that won't happen -- everything in the Compellent system is redundant, from the controllers to the paths to the disks," Hanrahan said. "Of course you can't ever say 'never,' which I suppose is why I still write to tape twice a day. But tape is a worst case scenario, and I would have to go to it with any SAN failure."
"With any kind of block movement or any data path activity, there is a risk of data loss, that's always a concern," said Dave Russell, research vice president for Stamford, Conn.-based Gartner Inc. "But from a pure risk perspective, that's no different from what a volume manager or even a file system is doing today, or any storage array."
Hands-off ILM a trend for other vendors, too
"Hitachi Data Systems Inc.'s Tiered Storage Manager software does similar movement but at the volume level," Russell said. "IBM's SAN file system moves data at the file level but not block level; its SVC [SAN Volume Controller] product moves data at the block level but not for this purpose."
"It would be difficult for most other storage vendors to implement this level of granularity because they don't monitor access at the block level," said Tony Asaro, senior analyst with the Enterprise Strategy Group.
So far, the closest comparable product is 3Par Data Inc.'s planned adaptive provisioning feature, announced in January and slated for addition to 3Par's InServ storage system next year. The product will allow users to preconstruct the data volume service level definitions they require (e.g. gold, silver and bronze) and automate performance within a given tier by load balancing disk resources.
EMC Corp. has plans to add frequency of access as a classification factor to its DiskXtender product, which handles policy-based migration between different tiers of storage, according to George Symons, EMC's chief technology officer of information management software. DiskXtender's next release, slated for later this year, will also automate more of the data migration processes once the customer defines the service levels for each tier of disk.
But, Symons said, EMC will stick with data classification according to several different factors, rather than simply frequency of access. "Just because you haven't touched a file in about six months doesn't mean that it's not important or that it doesn't need to be stored on more reliable, high-performance disk," he said. "And when should it be deleted? How should it be deleted? None of that is addressed by frequency of use."
Pricing and AvailabilityStorage Center QuickStart ILM is available immediately through Compellent's business partners. Pricing starts at $49,500 for a single-controller configuration that includes 6.4 TB of storage, automated tiered storage, thin provisioning and continuous snapshots. A second controller is estimated to cost approximately another $10,000. The system scales to over 300 TB on one platform.