AIX

Hello,

How can I know if ORACLE Database is running slow due to Memory or due to processing power ?

I have only Oracle Database running on a P4 with 4GB RAM.

Could anyone suggest any tools which can help me determine exactly if it is memory issue or processor issue.

Code:

vmstat -w 1 20

is a good start.

---------- Post updated at 03:47 PM ---------- Previous update was at 03:34 PM ----------

And I forgot to say: There are some threads in our AIX subforum about performance monitoring, tuning etc. Maybe you check those 1st and see if these help so one has not to write all the same again and again

This is an old machine running AIX 5.2 on P4

running an ORACLE Query and then running the command

sar 1 20

gives the idle time between 1 - 20

otherwise the idle time is 99

The memory is 4GB

How to know if it is processor issue or upgrading memory from 4GB to 16GB will speed up the ORACLE Queries.

As far as I know all the processing takes place in the memory.

Not sure what the query is. It can be a complicated or a trivial one. There are several kind of memory types Oracle is using, iiirc, and the biggest is or was the SGA (I am not up to date). Not sure if all those several types that might be configured for that instance can be satisfied with your 4 GB, not to forget that the OS needs some RAM too. Also it is a question on what disks your DB is placed. Slow internal SCSI disks or maybe fast and cached SAN disks.

I remember from DB2 a tool called db2expln with that you could easily check out if a query uses indexes, etc. or does slow table scans etc. I bet there is something similar for Oracle, ie. some kind of SQL analyzer. Maybe it is a missing Index etc...

If you do not mind please let the vmstat -w run while the query is running and post it here using code tags.

What might be also interessting could be following output:

Code:

lsdev -C -l aio0

If there is several problems, it always often a solution to turn on Concurrent I/O (mount option "cio" and in Oracle filesystemio_options=setall) for the data, redo and log filesystems only.
But maybe get the info up there 1st - maybe you find something suspicous.

Hi,

to correct zaxxon (sorry !) turn on EITHER cio (jfs2), dio (jfs) OR set in Oracle filesystemio_options=setall - never do both. If you have an older oracle version installed, it might not be able to handle the setall option - in that case choose async_io and go with cio or dio for the filesystems. You might want to consider smaller blocksizes for the redo logs too.

I would like to see the

Code:

vmstat -Iwt 2 30 ; vmstat -v ; vmstat -s

outputs from a busy time too - and I would like to know if you have jfs or jfs2 filesystems and if you have applied at least some basic tuning ?

Interesting would be what your database is doing - trading, reporting, both ... how many connections, these things. Keep in mind every connection to oracle takes memory - in certain cases dozens of MB ...

Kind regards
zxmaus

No problem at all, I am fallible I did not notice any problem from having set both though. Did you encounter/read about any troubles? I have some older guides that do not exclude each other way explicitly.

As you said and I read in the following link it is not needed anymore for Oracle >= v10 to mount filesystem with cio anymore:
http://www.ibmsystemsmag.com/aix/oct...y/21979p3.aspx

zaxxon,

we switch on cio or let oracle decide where to use cio because we want to avoid double buffering (what we have when both is on). If you switch off buffering completely (what you do with SETALL and cio both active) and you have a high transaction database, you are running the risk of saturating disks what used to happen in our company pretty frequently and slowed down the DBs more than not having anything activated.

But you are right - this is only true from Oracle 10g upwards - and we still do not know which DB version filosophizer is using

Kind regards
Nicki