On 7/4/2014 11:30 AM, Bosco Rama wrote:
Random thought: What OS & kernel are you running? Kernels between
3.2.x and 3.9.x were known to have IO scheduling issues. This was
highlighted most by the kernel in Ubuntu 12.04 (precise) as shown
here:
<http://www.postgresql.org/message-id/50BF9247.2010800@xxxxxxxxxxxxxxxx>
I'm on CentOS 6.4 which seems to be Linux version 2.6.32-431.20.3.el6.x86_64
But it is a VM, so disk I/O can be rather random as there are other
tenants. While improving performance is nice, I was most interested in
wy a pg_dump takes longer than a pg_restore (nearly 50% longer as it
takes about 2.75 hours to dump, but 2 hours to restore). It's
counter-intuitive as reading from a DB is usually faster than writing
into a DB. I think those LOs are getting me as our DB is LO-intensive
(most data is encrypted blobs: encrypted uploaded user files and
encrypted app-generated XML/HTML).
