Context: We are experiencing problems with large Solr indexes hosted on machines in EC2. One theory that's been offered is that this could be due to intermittent IO problems or spikes in IO latency.
Has anyone seen short (or long, I suppose) IO slowdowns on EC2?
A related question is whether anyone has seen any benchmarks of EBS compared to the "local" disk.
I know there's a hundred people jumping to their keyboards to tell me that IO on virtualized infrastructure sucks. Thanks, but my question is about whether it always sucks equally, or if sometimes it suddenly starts sucking more than usual.