RE: Physical RAM utilization

[goldfish]

Hi all,

I'm looking for a second opinion here (or more!).

We have an Intel PIII server running RH7.1. Initially this server had 4GB of ram with the standard RH7.1 kernel. Over a period of days it would eat all the available physical memory and then start to consume the swap space. One this happened the servers load average would shoot up over 50 and it would inevitably freeze.

As a test I removed 2GB of memory ... now it only took half the time to go belly up. Next I updated the kernel to the latest 2.4.18-27.7.x. This server is used as an Oracle database server which grabs a hefty 800Mb straight off. So after a reboot the system was using 1.2GB of RAM ..OK.. when our daily backup started on this server the memory usage went right up to use pretty much all available physical memory. At the moment the database is open but the system is not live. It currently looks like so:


12:48pm up 1 day, 20:38, 1 user, load average: 0.02, 0.02, 0.00

[root@rh71server01 /root]# cat /proc/meminfo
total: used: free: shared: buffers: cached:
Mem: 2114895872 2025791488 89104384 0 26578944 1858322432
Swap: 2097340416 0 2097340416
MemTotal: 2065328 kB
MemFree: 87016 kB
MemShared: 0 kB
Buffers: 25956 kB
Cached: 1814768 kB
SwapCached: 0 kB
Active: 1152692 kB
Inact_dirty: 483456 kB
Inact_clean: 222680 kB
Inact_target: 371764 kB
HighTotal: 1179584 kB
HighFree: 1024 kB
LowTotal: 885744 kB
LowFree: 85992 kB
SwapTotal: 2048184 kB
SwapFree: 2048184 kB
Committed_AS: 867460 kB


Having read a previous article top command memory question
in this forum answered in depth by Dolda I believe that I have nothing to worry about here ... however the seeds of doubt have been sown in my mind so I would like to hear from someone who has experienced a similar issue.

Also do I need to configure this kernel in any way to support a large memory config? If so what params do I need to change?

Any replies greatly appreciated!

Don't be surprised if I ask this question again. ...

Goldfish


[/url]

[andutt]

The kernels below 2.4.19 has documented troubles to manage large memory sizes.

Ivé had the samt problem running big servers on RH 7.3 with oracle with SAN solutions and 4gb RAM. When i changed kernel to 2.4.19 that dramaticly changed to the better, a couple of big fixes on memory and IO handling solved the most of our problems.

Try to upgrade to that kernel 2.4.19 at least and upgrade to RH 7.3 that i think is the most stable freeware distribution for servers. Otherwise it should not be a problem for memory sizes up to 4gb. Also enable hypertreading on you cpus for better scaleability for Oracles parallel querys.

Ive migrated about 60 unixserver to Linux that now runs RH 7.3 with 2.4.19 with oracle as a databasehandler, so if you need some more tip just post here or on my site http://www.utterberg.com/ under the forums section.

Regards

[Dolda2000]

Well, that's not all, though. I don't know for sure about the RH stock kernels, but I don't think that they come with High Memory support. Check your kernel configuration under Processor Type and Features, and you'll see the High Memory Support option. Set that to four GB and recompile. That should make your memory usage a lot better.
Also, when the swap started filling up, did you check what process takes it all?

[turls]

Red Hat 8.0 defininately has high mem support in kernel 2.4.18. This is set at 4GB.

I would also check with the Oracle boys aswell and raise a TAR. Could be some tuning issues around the DB etc. Is it using Apps?