I just took a couple of these for a quick spin so that I could have an answer if anybody asks about them tomorrow (I'm doing a LISA'13 talk about cloud storage performance). They really are pretty zippy. Basically they seem about 3x as fast as their predecessors, particularly with respect to disk I/O. The 1GB type is among the fastest in that category at over 15K (synchronous random 4KB) IOPS, for only 50% more per month than competitors. The 15GB type got over 28K IOPS, which is among the best I've seen in a public cloud (e.g. Storm on Demand 12GB or the astronomically priced AWS hi1.4xlarge).
Network-wise, I got the usual Rackspace throttling-induced asymmetry - 194Mb/s going from the 1GB instance to the 15GB, 874Mb/s the other way. Likewise, the cloud block storage significantly underperformed the instance storage at just under 8K IOPS - though that's still really nothing to sneeze at in a public cloud.
For something you can rent by the hour these are pretty sweet. It's not too hard to find 10KIOPS-capable machines elsewhere for much less, but then again those IOPS might not do you much good if the machines are network-constrained (and many of the cheaper providers do tend to be). As always, measure for yourself. It's always good to see that bar being raised.
Thanks for the feedback. I am the Engineering Manager for Cloud Block Storage and I'd love to get a few details of the tests you ran.
What region were you running the tests in and when did you conduct your testing? We very recently made some networking changes that should have dramatically improved the cloud block storage IOPS capabilities.
Thanks for your time and don't hesitate to reach out if you have any questions, comments or concerns.
Cool. Thought it was IAD, but wanted to verify we were talking about the same thing. We didn't officially launch the new flavors until 11/4 (even though they were out there before that) and we addressed some networking limitations just prior to launch. Thus, if you initially ran your tests with Cloud Block Storage prior to the changes, you very likely would have gotten the results you experienced.
In the performance tests we've run since the networking changes, we've seen substantially better IOPS performance with Cloud Block Storage. If you happen to rerun your tests we'd love to hear about the results!
The email announcing it it was rather vague. "A note from Lanham, Rackspace CEO, on a new definition and focus on performance."
It goes on to note how performance is important and invites clients to "contact a Racker to talk about your performance objectives.". The words "server" and "SSD" don't even appear in it. Well, at least he didn't invite us to "dialogue".
Network-wise, I got the usual Rackspace throttling-induced asymmetry - 194Mb/s going from the 1GB instance to the 15GB, 874Mb/s the other way. Likewise, the cloud block storage significantly underperformed the instance storage at just under 8K IOPS - though that's still really nothing to sneeze at in a public cloud.
For something you can rent by the hour these are pretty sweet. It's not too hard to find 10KIOPS-capable machines elsewhere for much less, but then again those IOPS might not do you much good if the machines are network-constrained (and many of the cheaper providers do tend to be). As always, measure for yourself. It's always good to see that bar being raised.