As i have already said: DESCRIBE YOUR SYSTEM!
Right now i have seen: 200G memory, 16 processors, running Oracle.
What i do not know (the list is not complete):
- which SAN / which multipath software?
- which OS level?
- which software? (that includes specifications, versions, etc.)
- what is the underlying managed system?
- how are your volume groups organised?
- network connections?
- is the system a cluster or not? (If yes, which one? HACMP? ORACLE RAC? else?)
Quote:
Originally Posted by
powerAIX
in iostat avg queue coloumn max time is 0.5 that is also not continuous.
I am at a loss what you mean. Try
iostat 1 and write that to a file for some time (say: one minute or so). You will have to filter out the lines with real devices because multipathing software (if you use such a thing) creates pseudo-hdisk-devices that show up in
iostat but are meaningless. For instance, with EMC PowerPath software you have one hdiskXX for every path and a hdiskpowerXX-device which is the real LUN. You have to watch only this.
Also in
netstat there are no crc errors please tell me what exactly i can check in that.
Well, CRC errors are your least concern, i'd say. More realistic scenarios are: misconfigured DNS servers so that name resolution takes long, the software running on externally authenticated user accounts (Kerberos, ...) which take a long time to authenticate and similar things.
Quote:
Originally Posted by
powerAIX
When checked with
< ps aux | head -1; ps aux | sort -rnk 6 | more> all are oracle processees and nothing else is running on the system. that is why am asking for help.
I
am trying to help you. I just can't if you don't give me some info. performance tuning is like tuning a car: i am trying to explain it to you, but so far i know it has so many hp power, but not even the model! And telling me something like "
when i step on the accelerator pedal it makes wwrrrrrmmmm" doesn't help me either.
Your ps-commands won't show you anything because AIX doesn't work like BSD. Instead, do the following:
First, look for shared memory that may be allocated. Use
ipcs -m and have a look there. If you want to have it analyzed by us: post it
complete.
Next, have a look where all your memory is spent. Try
ps -Alo pid,vsz,args (or variations, have a look at the man page of ps for details) to see all the processes with their allocated memory. Note that the memory unit used here is
pages (1 page = 4096 bytes), not bytes.
At last have a look at how the system is tuned: issue
vmo -a and look at the "minperm" and "maxperm" lines. Issue
vmstat -vs and look at the lines with "fs i/O with no pbuf" (or something such, i have no system at hand to look it up) and if there are high numbers there.
So, this is for a start. There might be more such requests, because performance tuning can be complicated. In every case: i suggest you look the commands up in the man pages and try to understand what is done with them and what the reason is for them being used. Because in the end the person most qualified to tune your system is: you! You know the system best and you sit in front of it.
I hope this helps.
bakunin