Raising the Roof - More Provisioned IOPS for EBS
aws.typepad.com
aws.typepad.com
I get that if you have a workload that needs 1000 IOPS and you only have 500IOPS QOS provisioned, then they will get queued and you will see latency. However, if you have a workload of 500IOPS needed and you have 1000IOPS QOS provisioned, what is the latency of that? Moreover, what is the variance of the latency of that?
Basically, if they are saying you get a 20 lane highway and you're guaranteed to get 200 cars a second across it, that doesn't mean that if you only have 10 cars that they're going to come through quickly. The quote they highlight talks about latency:
Amazon's Provisioned IOPS helped us improve our MongoDB query speed by over 60% and drastically reduced latency spikes to our clusters. We are thrilled to welcome the new 2000 P-IOPS volumes.
… but does the EBS PIOPS SLA talk about latency or variance in latency when you're below that number at all? Is it true that EBS really only experiences latency variation when you go over your provisioned IOPS?
Maybe I'm thinking about this the wrong way…
edited to add: After thinking about this some more I guess what I wanted to ask is: I have a database and I want to put it on EBS. If it doesn't even need 100 IOPS but I still want extremely low latency and latency variance, do AWS Provisioned IOPS guarantee this?
no.
closest thing to a citation from AWS here: https://forums.aws.amazon.com/message.jspa?messageID=370381#...
To answer the question added in the edit - no, PIOPS will not improve your database performance, as far as I can tell from the available information and my own benchmarks.