On Fri, 2004-08-06 at 14:56 -0400, Jan Wieck wrote:
>
> I have run it in a very small configuration (P3 667MHz, 640MB, single
> IDE) for 200 emulated browsers and 1000 items using Apache 1.3, PHP4 and
> MySQL 4.1.1 or PostgreSQL 7.4.2. The result is that they are head to
> head. Without pgpool MySQL is slightly better, with pgpool PostgreSQL
> pulls ahead.
>
> Two groups could possibly waste some time tweaking here, tune there, the
> tyical arm-wrestling of people who don't have anything better to do.
That would be me this weekend. ;-)
I'm working on getting this set up for testing postgresql/mysql on
various filesystems. (ext3 data=writeback, ordered, journal, reiserfs 3,
and possibly XFS.)
I have never tuned pgsql before and could use some guidance. I've read
the tuning guide from the web site, and the only values I have changed
from default are the shared_buffers and effective_cache_size.
The target system is:
AMD Athlon(tm) XP 2100+ uniprocessor
1GB PC133 memory
Fedora Core Rawhide
Kernel 2.6.7 (vendor supplied)
Hard drives: IDE /dev/hda=120GB 7200rpm, /dev/hdc=60GB, 7200 rpm
I've moved pg_xlog to /dev/hdc and the database resides on /dev/hda.
postgresql.conf is as follows:
max_connections = 100
shared_buffers = 7864 # 6% of main memory
effective_cache_size = 64000 # half of main memory
syslog = 2
syslog_facility = 'LOCAL0'
syslog_ident = 'postgres'
lc_messages = 'en_US.UTF-8' # locale for system error
message strings
lc_monetary = 'en_US.UTF-8' # locale for monetary formatting
lc_numeric = 'en_US.UTF-8' # locale for number formatting
lc_time = 'en_US.UTF-8' # locale for time formatting
The RBE will be running on an athlon thunderbird 1200 w/256MB DDR/266
over a 100mbit linksys switch. Server network card is a linksys (tulip
clone) and the network chipset for the client is (I believe) and rhine
of some sort. I don't have it set up yet and can't remember for sure.
(I'm concerned that the client may not be up to this task, particularly
the Rhine chipset. Should I be?)
Any comments or suggestions would be greatly appreciated.
Thanks,
Steve Bergman
