Re: How can I troubleshoot without informations ?

Just because the box has 2 gigs of ram doesn't mean that all that RAM is available to Oracle. Also I don't think that unless you have the enterprise version of NT (or whatever it is called) you can actually use all that RAM for 1 program.

In these "fuzzy situations", that is situations where you don't control what is going on you have to be skeptical about what people are telling you. I have found that people (especially technical people vs non-computer people ) have a tendency to "filter out" what you are telling them. They will assume things and so forth because they don't believe that the problem lies in such and such a place. (It may not, but you need to test it to make sure it does not.)

So I really would try what I suggest. Why? It is simple. It is non-destructive (shouldn't hurt anything). If what I suggest does not change anything you learn something from the fact that you tried it. If it does work then you have learned something.

Basically, I am skeptical that you are not running out of memory. I think you are or that you have reached a limit on the number of threads per process - there is one for NT.(each dedicated server process on NT is actually a thread not a separate program) It is not unusual for each dedicated server process to use 1 meg of ram not including sort area size so 800 connections would mean 800 megs + 500 for everything else means 1.3 gigs. Which is a lot of RAM.

So I think you have two possibilities

1. Running out of RAM
2. Running out of number of threads per process allowed under NT.
3. Some combination of 1 and 2.

You need to test these theories and see if they are true or false. (don't guess or assume, actually test them) In either event (if they are true or false ) you will have learned some valuable information.

Jim

"Dusan Bolek" <pagesflames_at_usa.net> wrote in message news:1e8276d6.0112112313.64aee951_at_posting.google.com...
> "Jim Kennedy" <kennedy-family_at_attbi.com> wrote in message
news:<0%rR7.8104$7y.73552_at_rwcrnsc54>...
> > It looks like you are using 235 Megs for the db block buffers (4096 *
60,000
> > /(1024*1024)) and 25 megs for SGA which used to be 30 megs. Also if
> > everyone is sorting at the same time you would have just under 200 megs
> > allocated to sorting. (800 * 258048/(1024*1024)) To me that adds up to
> > 235+30+200=465 say 500 megs of RAM being used by Oracle not counting the
RAM
> > allocated for each thread for other than sorting.
>
> Yes, that's true, but this box has two gigs of physical memory so
> there should be still free memory left.
>
> > They should be able to run perfmon on the NT box and monitor, free
memory, %
> > swap file usage. This should show how much memory is available. I
would
> > not decrease the SGA (25 megs is pretty small) Put it back to 30 megs
and
> > decrease the db_block_buffers by 35 megs. (51200).
>
> I believe that they can read a numbers from NT memory usage, so I
> think (better word is hope) that free memory is available.
>
> > Yes, and move the control files to seperate disks as you pointed out.
>
> It looks like there is no separate disk, just one big RAID5 set.
> That's stupid, but true. This is just some "small" server (just put
> here what doesn't fit anywhere), which grow too much. :-(
>
> --
> _________________________________________
>
> Dusan Bolek, Ing.
> Oracle team leader
>
> Note: pagesflames_at_usa.net has been cancelled due to changes (maybe we
> can call it an overture to bankruptcy) on that server. I'm still using
> this email to prevent SPAM. Maybe one day I will change it and have a
> proper mail even for news, but right now I can be reached by this
> email.
Received on Wed Dec 12 2001 - 02:08:44 CST