AIX

Currently server have load while there is no heavy things running, just oracle database/ application server oracle. I don't understand why server have heavy load, 22GB is under buffer, how to clean buffer/memory in AIX

Code:

load averages:  9.42,  9.43,  9.68;  05:25:08
141 processes: 125 idle, 16 running
CPU states: 98.3% idle,  0.9% user,  0.6% kernel,  0.0% wait
Memory: 30G total, 22G buf, 67M free
Swap: 18G total, 18G free

Moderator's Comments:

Please use [code][/code] tags instead of [quote][/quote] tags for code and samples

I can't see what the load is from that either, just that it's there. Try running top to see what's eating so much CPU, it could be intermittent. Or it could be lots of I/O.

I suspect buf is okay since you're not using any swap -- why shouldn't an OS use free memory as buffer, as long as it's willing to give it up when needed?

Dear Sir,

I have few questions,

1. i already paste the output of top in my thread.
2. how to check on AIX that i/o is taking load
3. how to clean buffer, is there any command that can fix buffer and memory.

Many thanks.

Buffer is to speed up things such as I/O. I don't think you need to worry about it. AIX manages Oracle loads very well. Are you experiencing any performance hit?

You only posted statistics, not anything relevant, i.e. the list of processes below it. We can see high load, but not which.

I'm not sure how to measure I/O on AIX, sorry.

Please consider what I said carefully. If your system was critically low on memory, wouldn't it be using swap? But it isn't. You've used no swap at all.

I'm not intimately familiar with AIX memory management, but we get a few posters a month panicking when Linux shows "0% free". Some of them also notice that 90% is in cache instead, and reboot their systems daily to "flush" it. But cache is as good as free -- the OS won't hoard it, it gives it away at need -- so their systems were perfectly fine all along, and there was nothing to panic about.

So I think 'buf' is as good as 'free', here. 22 gigabytes of good-as-free memory that your system can use at-will, but gets borrowed for cache when otherwise idle. Why not? It's wasted when sitting around doing nothing...

Please, post the output of vmstat -v.

You can view the I/O and the toll it takes on the machine with several tools which show different angles of the issue.

a) basics

Start with a concise description of your system: OS level, patch level, and the kernel parameters you use. AIX's memory management changed over versions and it is not clear in which direction to search for problems unless we have this information.

Code:

oslevel -g
vmo -o
schedo -o
ioo -i

b) even more basics

What is your system doing and how. OK, you said "Oracle DB", but: which version, are you using RAC? ASM? GPFS? Where are the filesystems from? (SAN? NAS? NFS? disks? rotating pancakes?) What else do we need to know to investigate? I might not have asked a complete questionaire but by now you should have gotten the gist of it.


c) collect data

Find out about the status of the asynchronous I/O (depending on OS level this is switched on - with different parameters - or not).

Post the output of (always as root):

Code:

vmstat 1
iostat 5 | grep -v ' 0.0'
svmon -G
vmstat -v

In case of the last command: issue it, then wait for 1 hour and issue it again. Post both outputs. The reason is: the "# of I/O with no fbuf" could be significant. Still, it is not the absolute value (which could come from a singular system state) that counts, but if it is relatively constant or climbing.

You might want to read the following post for more explanations about tracing errors:

https://www.unix.com/302363436-post8.html

I hope this helps.

bakunin

Corona is right. Memory management concept is almost similar in base level across the OSes.

I would not worry about high cache/buffer usage as long as I am not swapping a lot.

iostat and sar would provide very good information about I/O performance on AIX in a very similar way as Linux, Solaris, etc.