Hi,
First of all I must tell that my reality in a southern brazilian city is
way different than what we read in the list. I was lookig for ways to
find the HW bottleneck and saw a configuration like:
"we recently upgraded our dual Xeon Dell to a brand new Sun v40z with 4
opterons, 16GB of memory and MegaRAID with enough disks. OS is Debian
Sarge amd64, PostgreSQL is 8.0.3." on
(http://archives.postgresql.org/pgsql-performance/2005-07/msg00431.php)
Our old server was a very modest Dell Xeon 2.8 (512 Kb Cache), with 1 GB
RAM and one SCSI disc. This server runs PostgreSQL (8.1.4), Apache (PHP)
and other minor services. We managed to get a test machine, a HP Xeon
3.2 (2 MB cache), also with 1 GB RAM but 4 SCSI discs (in one sigle
array controller). They're organized in the following way:
disk 0: Linux Root
disk 1: Database Cluster
disk 2: pg_xlog
disk 3: a dir the suffers constant read/write operations
The database size stands around 10 GB. The new server has a better
performance than the old one, but sometimes it still stucks. We tried to
use a HP proprietary tool to monitor the server, and find out what is
the bottleneck, but it's been difficult to install it on Debian. The
tool is only certified for SuSe and RedHat. So we tried to use some
Linux tools to see what's going on, like vmstat and iostat. Are this
tools (vm and iostat) enough? Should we use something else? Is there any
specifical material about finding bottlenecks in Linux/PostgreSQL
machines? Is our disks design proper?
I really apologize for my lack of knowledge in this area, and for the
excessive number of questions in a single e-mail.
Best regards,
Alvaro