Dear list,
Our shop is running Solaris 2.6, Oracle database
8.0.6 and Veritas VxFS file system. At present, all database file access thru
the file system buffer cache, we decided to use the direct I/O mount option with
VxFS for performance reason. There is one issue I am not sure about and
The argument for direct io is that if you already have
a large cache (the oracle one), then its not much use
having a second copy of that cache (the unix one) -
that memory could possibly be better used elsewhere
(supporting more users, large sort sizes etc etc etc)
So having direct io is giving
Connor,
Thanks for your input. I agreed with you on what you said about the cache,
however what I was asking is the read-ahead that the file system provided,
for a physical read request, the file system actually read more blocks into
the buffer, so the next physical read request can be satisfied
My understanding is that most modern disks have track
buffers anyway which would obviate the need for
this...
(I'll happily stand corrected if anyone knows better)
hth
connor
--- KC [EMAIL PROTECTED] wrote: Connor,
Thanks for your input. I agreed with you on what you
said about the
