I'll add a few things (I've done a lot of SSD testing lately) as SSD performance is highly dependent on factors that don't influence hard disks: 0) Sequential

I'll add a few things (I've done a lot of SSD testing lately) as SSD performance is highly dependent on factors that don't influence hard disks: 0) Sequential performance is easy. If that's what you want, SSDs work pretty well and you can skip this post. The below points are for random IO. 1) Almost all SSDs are significantly slower at mixed read/write workloads (as a database would do) that either just reads or just writes. Sometimes as much as 1/4 the speed! 2) Random I/O throughput, especially for writes, is highly dependent on how full the disk is (assuming TRIM). For example, a 50% full is usually pretty fast, an 80% full disk is getting slower, and a 95% full disk is dog slow. 3) I have seen SSD controller and firmware version drastically impact performance. A recent firmware "upgrade" halved IOps on 100 of our SSDs (glad we tested that one...) 4) Time dependence! Many of my heavy I/O tests took 8+ hours to stabilize (often with very odd transition modes in between). Don't run a 30 second test and assume that's how fast the disk will stay. 5) Lastly, have many outstanding IOs in the queue if you want good IOps throughput. Think 32+. My recommendation overall: Test your actual application workload for 24 hours. Use a Vertex 4 with firmware 1.4 less than 75% full for your mixed read/write workload needs!