richardshen@canada.com wrote:[color=blue]
> We are having DB2 V8.1.5 installed on windows 2003 server has 2 intel cpu.
> I notice the CPU and Disk usage is very low no matter how busy is my application.
>
> How can we let DB2 make full use of system resource on Windows 2003?
> Thanks in advance.
> Richard[/color]
I do not run Windows anymore, and never ran DB2 on it.
I do run DB2 V8.1.5 on RHEL 3 ES, with two XEON processors 4 Ultra/320
SCSI 10,000rpm hard drives and two 100MHz ATA/EIDE hard drives (for the
little stuff in SMS tablespaces).
I used to run DB2 V6.1 on a machine with two 10,000rpm Ultra-2 LVD hard
drives and noticed that the machine was not compute limited (two 550 MHz
Pentium IIIs), nor was it IO limited (usually 1 megabyte/sec IO), but it
was slow anyway even when no other processes were using CPU. But you could
tell from all the clicking that the process(es) was seek limited. Hence
all the hard drives on the new machine.
I did not need all the storage capacity of all 6 spindles, but by dividing
things up better and using DMS containers (could not do that on RHL 7.3),
the new machine is 10x to 20x faster than before. It still runs at about
320% wait state (2 hyperthreaded XEON processors could go up to 400%),
which is frustrating. I am running 7 page fetchers and 7 page cleaners,
but the page cleaners all sit in state D (waiting for IO) a lot of the
time. Now maybe a wiser distribution of the tables over the tablespaces
would help. I cannot easily tell if it is seek limited, because the new
disk drives are silent.
What I have is the main data on three hard drives, and the index for those
data on the fourth. These are the fast SCSI drives.
Some medium sized stuff (for me): 3260 pages, total, are SMS on one of the
IDE drives and the tiny boring stuff 44 pages is on the other IDE drive
(where the log files also reside). I have studied it so that as I run it
the seeks should be minimized.
Since it is about 20x faster than the old machine (a >10-hour job runs in
about 35 minutes), I am not sweating it, but I wish I could cut the IO
wait time down since I run only about 40% of one CPU and the other 3.6
CPUs have little to do.
One bad idea is to run setiathome (entirely compute-limited). I run it
nice 19 which makes it least likely to get a shot at a processor, yet it
cuts the speed of DB2 in half. My present guess is that it is scrambling
the processor caches.
--
.~. Jean-David Beyer Registered Linux User 85642.
/V\ Registered Machine 241939.
/( )\ Shrewsbury, New Jersey
http://counter.li.org
^^-^^ 14:05:00 up 8 days, 21:20, 3 users, load average: 3.23, 3.21, 3.03
