Posted by: Beth Pariseau
Soon after filing a story on storage newcomer Atrato and its Self-Maintaining Array of Identical Disks (SAID) product, I started getting little pokes here and there from my analyst friends urging me to look at the story again.
The analysts had been reading stories from my esteemed fellow members of the Fourth Estate and noticed that some of the numbers weren’t matching up. Of course it’s always possible that reporters screw up, but when you have a good half-dozen or so reporting the story in slightly different ways, it could mean something different.
Take pricing, for example. I spent several minutes in my interview with CEO Dan McCormick trying to get him to tell me what the machine cost. The closest I could get was “six figures,” which seems to be what the Channel Register got, too. But they updated their article a little while later to add that Internetnews.com had apparently been given a number of $140,000 for 20 TB. Another source, HPCwire, had $150,000 for 20 TB.
I hadn’t focused on disk size in my article, but the specs reported there were inconsistent across news sources too: HPCWire and CIO Today reported the system uses 2.5-inch disks, but Byte and Switch reported 1.8-inch disks. I had assumed 3.5-inch disks, but StorageMojo’s Robin Harris pointed out that the kind of density Atrato’s talking about would be impossible in a 3RU system using 3.5-inch disks.
But that’s if the system is 3RU. Atrato’s own website doesn’t get this straight, either. The solutions section quotes a spec of “over 3000 data streams in 5RU” (click on “streaming” on the right hand side of the flash object at the top of the page), while the products section specifies “3600 data streams in 3RU.” Harris was given a preview of the product a few months ago, and was originally told the box was 5RU.
In fairness, there are areas where the nature of Atrato’s product makes the kind of specs we in the storage industry are used to seeing tricky. Because the system allows customers to throttle parity, the capacity stats get a little complicated. Most of the news sources I saw either reported the same total raw capacity number, 50 TB, or got into different permutations of how the capacity is distributed according to your reserve space for parity protection.
On the IOPS front, what I found was actually fairly consistent, either “over 11,000″ or the exact number, “11,500.” The one place I saw a major discrepancy was in the details about the SRC deployment at an unnamed government customer, which claimed 20,000 sustained IOPS. Their explanation for this is that 11,000 to 11,500 is where they’re quoting to be safe but that the 20,000 at the SRC customer represents the fastest speed they’ve seen in the field on a carefully tuned application.
But Harris took issue with the “11,500″ number, saying it’s too specific to really mean much, since IOPS are dependent on a number of factors.
“One possible take [on the discrepancies] would be, how many of these things have they built?” Harris pointed out. “With contract manufacturing, you don’t start building until you get volume, and you don’t get volume until you start convincing customers you’ve got something.” In this chicken-and-egg cycle, it could be that some of the Atrato arrays shipped to date have been 5 RU and they’ve decided to make more in the 3RU size. “But either way, they should get it straight on their own website,” he said.
Atrato got it straight in the press release announcing the product, identifying it as a 3U device, twice. No mention was made of a 5U box.
Also frustrating the propeller-headed among us is the lack of in-depth technical detail about how the product exactly works from Atrato execs. They wouldn’t tell any journalists exactly what disk errors their software claims to fix and other items details such as how the product connects to servers–iSCSI, NAS, or FC?
It appears Atrato has at least one potential customer commenting on this over on Robin’s blog:
It’s a pity that there doesn’t seem to be anything on their site about the connectivity options, processor redundancies, replication or clustering. If they provided a way to create a cloud of these they would probably be on the top of my solution list for permanent near-line archiving of about 60TB of data.
And it would be a pity, if Atrato really is sitting on something truly revolutionary, but the message just isn’t getting out there.
Then again, I’m writing about them again, aren’t I? “The conspiracy side of my brain tells me they could also be doing this to get maximum press,” Harris added. In that case, I guess we’ll find out if there really is no such thing as bad publicity.