At 04:45 PM 8/21/2006, Marty Jia wrote:
>I'm exhausted to try all performance tuning ideas, like following
>parameters
>
>shared_buffers
>fsync
>max_fsm_pages
>max_connections
>shared_buffers
>work_mem
>max_fsm_pages
>effective_cache_size
>random_page_cost
All of this comes =after= the Get the Correct HW (1) & OS (2)
steps. You are putting the cart before the horse.
>I believe all above have right size and values, but I just can not get
>higher tps more than 300 testd by pgbench
300tps on what HW? and under what pattern of IO load?
300tps of OLTP on a small number of non-Raptor 10K rpm HD's may
actually be decent performance.
300tps on a 24 HD RAID 10 based on Raptors or 15Krpm HDs and working
through a HW RAID controller w/ >= 1GB of BB cache is likely to be poor.
>Here is our hardware
>
>
>Dual Intel Xeon 2.8GHz
>6GB RAM
Modest CPU and RAM for a DB server now-a-days. In particular, the
more DB you can keep in RAM the better.
And you have said nothing about the most importance HW when talking
about tps: What Does Your HD Subsystem Look Like?
.
>Linux 2.4 kernel
>RedHat Enterprise Linux AS 3
Upgrade to a 2.6 based kernel and examine your RHEL-AS3 install with
a close eye to trimming the fat you do not need from it. Cent-OS ot
Ca-Os may be better distro choices.
>200GB for PGDATA on 3Par, ext3
>50GB for WAL on 3Par, ext3
Put WAL on ext2. Experiment with ext3, jfs, reiserfs, and XFS for pgdata.
Take a =close= look at the exact HW specs of your 3par.to make sure
that you are not attempting the impossible with that HW.
"3par" is marketing fluff. We need HD specs and RAID subsystem config data.
>With PostgreSql 8.1.4
>
>We don't have i/o bottle neck.
Prove it. Where are the numbers that back up your assertion and how
did you get them?
>Whatelse I can try to better tps? Someone told me I can should get tps
>over 1500, it is hard to believe.
Did they claim your exact HW could get 1500tps? Your exact HW+OS+pg
version+app SW? Some subset of those 4 variables?
Performance claims are easy to make. =Valid= performance claims are
tougher since they have to be much more constrained and descriptive.
Ron
