What I don't understand is on their flagship product, an all SSD drive array, they claim 500,000 IOPS (sustained I/O rate), yet i have heard of many people doing more than 1 million IP's/s in http and in database. If others can do a million how are they if something that is seemingly as fast as can be with an all SSD system can only do half that at a pure I/O level no less.
I think 4kb blocks is the biggest design fault and performance killer for databases. They claim to store blocks serially if IO size is larger than 4kb but in practice I haven't seen it working very well.
Yes, but it should be noted that to see this speed up you can only do read caching. You can do write buffering in RAM but you need to mirror it to another node or you risk data loss in a power outage situation. RAM-over-network is roughly equivalent to SSD performance. It should also be noted that not all SSDs are created equal. You can get an SSD that will do tens of thousands of IOPS or you can get SSDs that do over a million IOPs per drive.
