The Memtech solid-state drives Trent pointed out above seem to resolve these
issues fairly well. Up to 60GB, Average access .1ms, 8M cycles, 8 year
warranty seem to nail those concerns down (though STR seems to be left out of
the spec). Shock rating of 2000Gs is rather impressive too. You forgot cost
in your list though. ;-)
With a .1ms average access, these make 15K RPM SCSI look turtle-slow.
Is there a need for solid-state drives for servers?
It's application-driven: for example, if you really want tps to scream, use an
SSD for the database "after-image log". This file captures every write to the
database - and is the commit point for releasing file locks. As tps systems
are intrinsically synchronous, until that after-image file is updated with a
committing transaction, the entire system comes to a screeching stop, with a
whole system full of users waiting for their next screen refresh.
An alternative is to stick a non-volatile, secure write cache in the path from
the tps to the after-image log disk, and run something like Legato's
"PrestoServe" on it. Writes get captured by the secure write cache and the
file locks are released at that point, instead of waiting for the same write
to land on a cylinder. Plus, writes can be gathered at the cache and then
blasted to the spindle in block writes, instead of lots of individual write
transactions. As tps transactions are usually quite tiny in size, this cuts
average latency *and* queue depths down quite effectively...and results in
much happier users...
/daytripper (been there, designed all that, and thousands of AlphaServers
shipped with the secure write cache and a licensed instance of PrestoServe)
