High WAIT IO

[Ashwin_4]

Hi,
We have oracle 8174 database running on HP L-class server 11.x 4 processor and 1Gb server. When we run the batch billing process, sar -u show high %wio and all 4 dbwr process consume high CPU (dbwr process appears in top command).
23:05:49 %usr %sys %wio %idle
23:05:53 5 2 32 61
23:05:57 7 2 23 68
23:06:01 5 1 34 60
23:06:05 5 2 43 49

Pl help.

[Sridhar Bhaskarla]

Hi Ashwin,

Also post your 'sar -d 2 5' output. We can only tell that the system is waiting on the IO from the output you attached. Also, post your VG and LV configuration such as how big are you disks, how many VGs, how many disks are in each VG, how many LVs in each VG, how big are LVs etc.,

-Sri

[Steven E. Protter]

Oracles general guidelines call for raid 1 or raid 10 storage for redo logs and data. If you are sitting on a busy raid 5 array, that could explain high wait times, especially if you are sharing your disks with other servers.

The sar -d ouptug should show wait times of 6 or higher which shows a serious i/o problem. Please tell me what the raid setup is for your oracle database.

Also note that Oracle 8174 is obsolete and significant performance enhances are in the current 9.2.0.4 release of the database server, though that uses more memory.

SEP

[Ashwin_4]

Hi,
15:30:18 device %busy avque r+w/s blks/s avwait avserv
15:30:20 c1t2d0 2.00 0.50 2 28 1.56 15.87
c2t2d0 1.00 0.50 2 24 2.45 9.34
15:30:22 c1t2d0 2.00 0.50 2 20 2.97 14.31
c2t2d0 1.00 0.50 1 16 5.74 11.55
15:30:24 c1t2d0 3.52 0.50 4 44 1.05 19.85
c2t2d0 2.51 0.50 3 40 1.35 12.14
15:30:26 c1t2d0 2.99 0.50 4 41 5.24 19.60
c2t2d0 2.49 0.50 3 37 6.29 26.18
c4t1d0 0.50 0.50 1 16 5.48 6.67
c5t0d1 0.50 0.50 1 24 2.52 5.09
15:30:28 c1t2d0 4.00 0.50 6 56 5.09 14.98
c2t2d0 3.50 0.50 5 52 5.80 14.64
c4t1d0 6.50 0.50 5 80 5.16 12.38
c5t0d1 6.50 0.50 6 97 3.90 10.01

Average c1t2d0 2.90 0.50 4 38 3.61 17.00
Average c2t2d0 2.10 0.50 3 34 4.66 15.97
Average c4t1d0 1.40 0.50 1 19 5.22 11.42
Average c5t0d1 1.40 0.50 2 24 3.64 9.09

Current the batch process is not running server is almost idle. We are having HP-UX AUTO RAID. One VG with 17 LV's each of 5Gb.

[Tim D Fulford]

Hi

this is more of a comment than an answer to the query..

Whilst the above advice is correct (disks may be the problem). The sar -u is for 4 second averages... I agree they are consistent but the total sample is some 16 seconds. Ths is not entirely meaningful as the server could be in the middle of a highly disks intensive activity. In the future it would be better to post 1 or 5 minute averages.

regards

Tim

[Alzhy]

I suppose if you've also provided us with vmstat output - it will show that processes will also be blocked for IO.

With Autoraids. this will be "normal". WaitIO's combined with individual disk queuing and process blocked for I/O will (99% of the time) mean simply that your I/O subsystem is not able to handle all the requests for I/O. While this may be worrying -- it will all depend on your application(s). If the desired result and response times are still being met - then leave it as is. If this is a new occurence then you will probably need to look at what new processes were intoroduced or if you're running double or triple what should be running.

WaitIOs in the past were always a big worry to me but since I've started seeing them even on high end arrays like the EVA.. I worry less anymore as long as my response times are good and my batches finish within reasonable times.