[ltp] SSDs and AHCI?

[Theodore Tso]

On Tue, Jun 23, 2009 at 06:02:20PM +0100, Richard Neill wrote:
> Do you have any details on AHCI doing harm to the OCZ Core SSDs? All I  
> can find are references that say it won't help (and it confuses 
> benchmarks).

It was NCQ that could cause performance problems for the OCZ Core 2,
but it looks like the OCZ Core 2 has been blacklisted already:

http://markmail.org/message/jvjpmcdqjwrmyl4w

> How much do write latencies matter? Doesn't Linux batch them up anyway  
> (especially in laptop-mode?)

Linux normally tries to do patch things up, but there are times when
it has to wait for a write toc complete; for example, if the
application uses the fsync() call, that's an explicit request (for
safety reasons) not to return until the data is safely stored on
stable storage.  For example, mailers use this quite commonly; so do
various database programs/libraries.

The write latencies also translate to slower random read/write
throughput numbers.  This may or may not matter, depending on whether
your application is disk I/O bound.  This should be obvious; if every
so often, the disk is going to not respond for a full second or two
while handling a random 4k write, and it can't handle multiple writes
in parallel, it's going to make your average MB/s numbers look pretty
ugly.

						- Ted