We recently did some tests with SSD because write response time of TEMPDB was below expectations.
Our first test was very promising. Our 224 GB SSD disk was used to store 12 x 1GB TEMPDB files on. Read as well as write response times were well below 5ms/transfer.
In our enthusiasm we also added our most intensively used user database mdf file on it, filling the disks to 95%. Here things went wrong. Read performance was still great but write performance decreased until 200ms/transfer.
After some reading – SSD should not be filled to its full capacity – we removed the large mdf file. But to our surprise, write performance was still unacceptable.
Apparently this is standard behavior when you put write intensive data files (e.g. TEMPDB) on SSD disk behind a HW RAID controller.
Anyway, for us SSD is NOGO. These things aren’t cheap by default, if you can only use 50% of its capacity, discussion was closed.
More info here; http://www.sqlservercentral.com/blogs/hugo/archive/2011/06/08/notes-from-solid-state-for-sql-server-with-wes-brown-from-today-s-idera-broadcast-ssd-best-practices-for-database-servers.aspx