Re: Oracle Eats RAM, none left for Gods

In article <7g2ame$r98$1_at_nnrp1.dejanews.com>,   JoshNarins <joshnarins_at_my-dejanews.com> wrote:

> We had 100% RAM filled ( constantly 50-100M in swap ) and horrid performance
> with Oracle8 and Solaris2.6 with 3G of RAM.
>
> We added a GIG of RAM.
>
> We STILL have 50-100M in swap.

This is OK (as Thomas T. Kyte already pointed out). This is just allocated, the system is not swapping.

>
> What configuration setting is allowing Oracle to consume all available memory?
> OR, since we aren't running anything else on the machine, is using all memory
> this way better than limiting what this greedy monster consumes.

This is not only Oracle, also the Filesystem is using memory (and it is using a lot!)

>
> Performace is better with the new RAM, but I still think Oracle's behaviour
> is wrong. We used to page out under heavy stress. We haven't seen loads like
> before yet (we saw mid 20s) and no one has complained yet withg the new RAM
> HOWEVER
>
> an answer exists.
>
> Thank you in advance for any assistance in providing it,

I (we) have also seen this on a server of a customer (a big financial institute here in Switzerland) in a similar setting (OK, it was still Solaris 2.5.1 when we encountered the problem but at the end we had 5GB RAM in the machine). The problem seems to be that Solaris with maybe 3GB or more RAM and Oracle with a SGA >~1GB don't like each other; it even seems they are hating each other. Our conclusion was, that this is somehow Filesystem-Cache related. If you run 'vmstat' (the first line returned is just the average since startup, so don't look at it), there is a column "sr" which is the "memory page scan rate". On heavy loaded systems this may go up to 200. In the above combination it may show you scan rates of 7000 to 10000. Not a healthy system. One solution can be to use raw devices as data files for Oracle. We decided to upgrade to Solaris 2.6 (you are already there). In Solaris 2.6 you can bypass the Filesystem-Cache. This was what we did and suddenly the scan rate went down to ~100 and this performance problem was gone (We still had another one from a silently activated trigger, but this is another story). Mounting a filesystem (under Solaris) without caching it you do with the option 'forcedirectio' (you can do this even on the running system with e.g. 'mount -o forcedirectio,remount /dev/rdsk/c0t0d0s5 /oradata'). May be this will also solve your problem?

Leo Mannhart

Oracle Consultant                      http://www.trivadis.com


-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own