DISK busy 100% system slow

[prash_1]

Dear Gurus,

I am using HP-UX 11.11 in rp3410 with 2 CPU.

Disk Station DS2300.
oracle 9.20 and sap installed in the system.


In the system two internal Disk c2t0d0 used only for OS. and mirred with c2t1d0.

System is very slow because of internal disk showing 100% busy.
Here I am attach O/P of sar

[Pete Randall · Accepted solution]

I wonder if you're swapping. Is your swap device located on the internal disks? What does "vmstat 1 5" show in the "po" column?


Pete

[Steven E. Protter]

Shalom,

If your disk is busy 100% the most likely cause is swapping. I'd suggest vmstat or swapinfo -tam

http://www.hpux.ws/system.perf.sh to monitor in the background.

I'm sure that if oracle is hosting files on local disk and the environment is write intensive this could also be a contributing factor.

SEP

[prash_1]

Thanks for reply.
Yes in the system internal Disk is used for swap
it is /dev/vg00/lvol2.

TYPE AVAIL USED FREE USED
dev 21004288 4869624 16134664 23%


#vmstat 1 5
pi po
37 91
12 698
46 694
66 768
41 782
26 842.

[Pete Randall]

Wow, with those "po" numbers, I'm amazed that the machine is moving at all. You need to reduce the load on the machine or add memory. How much RAM do you have?


Pete

[prash_1]

Thanks,

In system 4GB ram is there

#swapinfo -m
TYPE AVAIL USED FREE USED

memory 3081 570 2511 19%

reserve - 4871 -4871

this will Happen some times only
and If I restart system it will be normal.
but I can't take restart because it is production system

[Pete Randall]

You probably aren't getting much production out of this "production" machine!! I would weigh the improved performance that results from a re-boot against the poor performance you're experiencing now and then decide if you might not be better off re-booting.

In the meantime, are there any unnecessary processes that you can kill off?


Pete

[Patrick Wallek]

It appears that you need A LOT more RAM on this machine. How much do you have? Whatever the amount, I would add at least 8 GB since you are using almost 5GB of swap space.

[prash_1]

thanks,

my sap users are loged in so I cant't take
restart.

How can I find which are unusable processes.

[Patrick Wallek]

>>>How can I find which are unusable processes.

That is not something that we can tell you. YOU must look at your 'ps -ef' output and try to decide what is or is critical.

[prash_1]

sorry,
I want to say some processes which are initiated by oracle & some by SAP
in that processes I am not sure which is
Necessary

[Patrick Wallek]

Again, that's not necessarily something we can answer. You must talk to your Oracle DBA and/or SAP Admin and see what they say.

[Pete Randall]

It might be helpful to look at a sorted list of the top memory usage processes:

UNIX95= ps -eo vsz,pid,args | sort -rn | more

If you can find something unnecessary near the top of that list and eliminate it, you may see some improvement.


Pete

[Steven E. Protter]

Shalom,

If you have zombie processes, which are disconnected from their parent, these processes could be doing all kinds of useless i/o, slowing down the system without providing any useful data.


If you identify a lot of zombie processes, the best way to fix it is to arrange a short system outage and stop/start sap.

There are several other scenarios that could cause a lot of needless disk i/o. Users not terminating their sessions properly, issues like that.

The key is to identify and possibly kill individual processes that you know are hammering performance and are not providing useful results/data. This may annoy the users a bit, so tread carefully.

SEP

[James R. Ferguson]

Hi:

> How can I find which are unusable processes?

If you have to ask that, you are the *wrong* person to decide! To begin killing processes without any understanding of what they do will certainly create a worse mess. Of course, it will likely force you to reboot which may then "fix" your immediate problem.

I urge you to consult your DBA to ascertain the amount of memory devoted to SAP. Perhaps its buffer pool is too large.

I also suggest that you make sure that your Unix buffer cache isn't set to some very high value. Your values for 'dbc_max_pct' and 'dbc_min_pct' should probably not exceed 5% under your conditions.

Regards!

...JRF...

[prash_1]

Thanks to all. I am consulting with my DBA administrator for oracle parameters & processes.

[prash_1]

Thanks.