How to identify if there is a memory bottleneck?

[Zhang Zhao gui]

From GlancePlus, I can see that most of the time especially during the working
hours the memory usage is
between 99% and 100% and the swap Utilization is about
63% to 70%. But the CPU usage is not high (about 60% to 80%). Disk usage is
very low(10-20%).
Am I getting short of memory or if there is a memory bottleneck?

[Dan Hull]

It sounds like a memory shortage. Before you buy more memory though, check your
kernel parameter called "dbc_max_pct". The default value is 50%, but this is
for small memory systems. If you have 256M of RAM or more, consider reducing it
to around 25%. This may free up some RAM for other uses. If you have 1G or
more, try reducing it to 10%.

[Zhang Zhao gui]

dbc_max_pct is maximum dynamic buffer cache size as percent of system RAM
size. Our system got 896MB RAM.
Currently the buffer cache size is about 430MB, I wonder if I followed your
suggestion and reduced it to 20% of 896MB, which is only about 200MB. Will
there be any side effect such as slower system response or
not enough buffer cache thus causing other problems?
Is this HP's official recommendation for solving such
kind of performance problem? Are there any guidelines
provided by HP which we can follow?

[R Madhavan_1]

Yes, you are right, reducing the I/O buffer may increase your I/O and affect
your system performance in another way.
But for memory bottleneck, you may probably want to look at your page in/out
and the memory queue over a period of time. (use glance or vmstat). If both are
high, it may be a indication of memory pressure in your system.

[Chen Khin wai]

Another way to determine whether you have overconfigure your buffer cache is
issue "sar -b" (specify your own interval)

If your %rcache and %wcache is 100% (means a read and write is succesful in
getting it from the cache) then try to reduce it a little. If after reducing
and its still 100%, means you are actually overconfigure your buffer cache as
most of the read & write are still in the cache.

If %rcache and %wcache is less than 99%, means disk I/O is involved and there's
not enough buffer cache.

[Zhang Zhao gui]

I run "sar -b 5 10" and found that the on the average
%rcache is 99, wcache is 68. How do you think of my
buffercache? From GlancePlus, I noticed Swap Util is
about 65% to 68%, does it show that there is a lot of swapping or paging? I
show you what I got from Glance as below, do you think if there is any memory
bottleneck based on the following data?


B3692A GlancePlus C.02.30.00 12:17:46 isdux3 9000/802 Current Avg
High
--------------------------------------------------------------------------------
CPU Util SN N | 40% 35%
87%
Disk Util FF | 4% 64%
89%
Mem Util S SU UB B | 99% 99%
99%
Swap Util U UR R | 67% 67%
67%
--------------------------------------------------------------------------------
MEMORY REPORT Users= 6
Event Current Cumulative Current Rate Cum Rate High Rate
--------------------------------------------------------------------------------
Page Faults 333 17842 40.6 92.2 1085.4
Paging Requests 83 8428 10.1 43.5 581.8
KB Paged In 0kb 1.2mb 0.0 6.2 98.4
KB Paged Out 0kb 0kb 0.0 0.0 0.0
Reactivations 10 195 1.2 1.0 3.8
Deactivations 10 195 1.2 1.0 1.9
KB Reactivated 0kb 0kb 0.0 0.0 0.0
KB Deactivated 0kb 0kb 0.0 0.0 0.0
VM Reads 0 91 0.0 0.4 4.2
VM Writes 0 0 0.0 0.0 0.0

Total VM : 366.2mb Sys Mem : 79.4mb User Mem: 413.0mb Phys Mem: 896.0mb
Active VM: 83.2mb Buf Cache: 392.4mb Free Mem: 11.3mb Page 1 of 1
ProcList CPU Rpt Mem Rpt Disk Rpt NextKeys SlctProc Help Exit