Operating System - HP-UX
1748015 Members
4577 Online
108757 Solutions
New Discussion юеВ

Oracle I/O related problems

 
Ramalaxma
New Member

Oracle I/O related problems

Our databases are extremely I/O (write intensive) bound.
I see in global waits on glance (semaphore 92% blocked on IO 1.9% ). What does that mean? We tried all possible things from database side. Is there anything I can do from unix side?
Thanks for your info. in advance.
7 REPLIES 7
Ted Ellis_2
Honored Contributor

Re: Oracle I/O related problems

what kind of hardware is the database sitting on? Fibre or scsi? RAID configuration, etc.

Ted
Ramalaxma
New Member

Re: Oracle I/O related problems

Thanks for the reply.
Database is on: L class server(4cpu,16GB RAM);Uses EMC storage with SRDF(synchronous); Uses raw devices; Produces 30GB of redo everyday.
We see hell lot of "logfile sync" wait events in database.
We do not use Async IO.
Any pointers to where/what to look at?
George A Bodnar
Trusted Contributor

Re: Oracle I/O related problems

You are probably waiting on the SRDF stuff to sync since this is Sync mode. Basically you are probably stuck on very slow I/Os based on the error messages.

You should try to look at the performance on the Symmetrix (specifically the SRDF portion) to see if that is your problem. Do you have WorkLoad Analyzer that you can use to get statistics from the Symmetrix?
George A Bodnar
Trusted Contributor

Re: Oracle I/O related problems

Also you could look at sar -d and check the avwait/avserv times to see what your response times are on the disks in question. You can also get this data from Measureware.
Sridhar Bhaskarla
Honored Contributor

Re: Oracle I/O related problems

Hi,

Waiting on semaphore indicates that there are hot tables in the database and log sync as well. Try to adjust them on different disks. Also your SRDF background event is also contributing it.

EMC does not recommend, but we are doing striping of the logical volumes on XPs here. It seems to yield significant performance increase. Particularly you may want to try to stripe your redo log lvols. Though the LUNs (volumes) are striped in side the disk system, lvol striping will further improve performance if you chose your stripe based on the stripe of RAID.

Did you try using file systems for the database?. With online JFS you get the mincache=direct and convosync=direct so that you will get the same raw functionality. File systems are more manageable from the sysadmin's perspective.

-Sri
You may be disappointed if you fail, but you are doomed if you don't try
Volker Borowski
Honored Contributor

Re: Oracle I/O related problems

Hi,

with 30GB of redo each day, how big are your online relogfiles ?
Each logswitch will force a checkpoint, so if you have small redologs, your databasebuffer will be flushed to disk all the time. with 30GB of redo each day, I'd recommend to have 200MB-500MB online redologfiles and may be up to 6 groups.
This will give you 60-150 logswitches a day, which might be still too many.

Hope this helps
Volker
Ramalaxma
New Member

Re: Oracle I/O related problems

Thanks to all of you for your
suggestions. Some more details about my databases.
redolog size is 500Mb, there are 15 memebers. It switches a log every 7 min. in peak time. Implemented SAME methodology (stripe and mirror everything) stripe size 64kb. planning to go for 1MB. We know SRDF is a bottleneck. Moving to latest series. If you guys can help me to speed up lgwr I will be happy. log_buffers=1MB. Our problems are redolog,lgwr oriented. There is no way to use parallelism in lgwr. what else I have to do?