>  Quantum is the only manufacturer to offer true zero latency reads (they did
>  their cache right).
>
>I always thought that disk drive based caches were counter-productive if
>you're running an os with its own disk cache, as NetBSD has.  Is that wrong?
>Has anyone ever measured the effect of a drive based cache on BSD
>performance?

Disk caches can make a large difference.  In the case of zero latency reads, 
the drive starts pulling the first piece of data on a cylinder that is
part of your request that it finds from its current head position, offsets
it into the cache, and  fills in the "begginning" portion of your request
when it gets there.  If you can't do this, you must incur a larger rotational
delay to get to the beginning of the request before you can start pulling data.
The drive will also do read ahead while its waiting for more requests.

Controller caches, on the other hand, are a subject of much debate.
--
Justin T. Gibbs
===========================================
  FreeBSD: Turning PCs into workstations
===========================================