Re: Datafile buffering w/ OSF1

Antti Jarvinen wrote:

> I'm having a slight swapping problem with digital unix 4.0 and oracle804:
> bash-2.02$ vmstat 1
> Virtual Memory Statistics: (pagesize = 8192)
> procs memory pages intr cpu
> r w u act free wire fault cow zero react pin pout in sy cs us sy id
> 2141 33 12K 62 50K 782M 25M 212M 134M 67M 15M 158 1K 747 38 5 57
> 3141 32 12K 50 50K 1336 91 549 3321 252 70 387 1K 1K 26 16 58
> 4134 36 12K 42 50K 982 54 377 58 225 68 354 2K 1K 40 23 37
> ...
> and almost all processes (including ora*) are regularly swapped out and in
> causing pretty ugly slowdown.

How much RAM do you have? Oracle recommends a minimum of 128 MB. Do you have 500 MB? Check kernal settings (or in sysconfigtab) you need to have shared memory max size
set to 2GB - 8MB (1.992 GB) and segments to 32. Unless you can do this you will get nowhere. What percentage of total RAM is the SGA, you should aim for about 60%
What are your applications needs? Can you put Oracle on a separate box from the app - this is recommended practice...
Luck,

>
>
> I'm thinking this might be a file system buffering problem. Our application
> for most part does things periodically: there are processes that do things,
> then go to sleep() for some time, wake up and begin stressing the DB again.
> Every wakeup seems to cause swapping. Because this application does not use
> files, database is the storage for all necessary stuff then I'm not sure where
> the file access are coming from and therefore:
>
> - does 8.0.4.3.0 on 4.0d open datafiles in such a mode that the filesystem
> does not cache the datafile contents again?
>
> As this uses datafiles in AdvFS-disks, no raw partitions are used.
>
> Also, I wonder if it does any good to lock SGA into memory as it seems
> to be possible. It should prevent the system from swapping the SGA but
> then it would increase swapping of other processes? Vmstat -P shows the
> following breakdown from "wired" part of memory:
> WIRED Pages Break Down:
>
> vm wired pages = 1199
> ubc wired pages = 4652
> meta data pages = 1958
> malloc pages = 41320
> contig pages = 279
> user ptepages = 652
> kernel ptepages = 54
> free ptepages = 4
>
> .. total system memory is only 64K pages then what the fsck is this
> "malloc pages" or how do I find out what is eating it? It is clearly
> not SGA as I see ora* -processes with good sized being swapped.
>
> Thanks for your help,
>
Received on Mon May 08 2000 - 00:00:00 CDT