RE: Bufwaits
Art,
I have an older document I pulled from the archives, it was dated 5/12/98.
At that point you recommended < 3%, has something come up that changed your
thinking to < 5-7%. I only ask this because bufwaits are so lightly
documented in any Informix docs, ours are about 4%, and of course I've been
attempting to get them below 3%.
TIA,
John E. Porcello, Jr.
Corporate DBA
Kaman Corporation
jep-corp@kaman.com
My views and opinions may not reflect those of Kaman Corporation.
-----Original Message-----
From: Art S. Kagel [mailto:kagel@bloomberg.net]
Sent: Tuesday, March 23, 1999 3:14 PM
To: informix-list@iiug.org
Subject: Re: Bufwaits
Luse Harold wrote:
>
> Art (and others),
> Here is the onstat -p. All comments are appreciated. Thanks.
>
> Informix Dynamic Server Version 7.30.UC5 -- On-Line (Prim) -- Up 5 days
> 22:21:14 -- 81040 Kbytes
>
> Profile
> dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached
> 8393998 3901487 114512135 92.67 401142 725852 2207625 81.83
>
> isamtot open start read write rewrite delete commit
> rollbk
> 82285018 2283251 7423619 50162223 1396849 41521 11344 73663 80
>
> gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs
> 0 0 0 0 0 0 0
>
> ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes
> 0 0 4 14495.97 2968.34 1343 3428
>
> bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans
> 1038556 0 234209482 0 0 316 6811 172645
>
> ixda-RA idx-RA da-RA RA-pgsused lchwaits
> 1017321 170911 4922438 6076312 14036
wait% = (bufwaits / (pagreads + bufwrits)) * 100
= (1038556 / (3901487 + 2207625)) * 100 = 17%
Your bufwaits % is 17% which is a bit high (I like to see 5-7%) there
may not be anything to worry about but you can try to bring it down.
You may try increasing the number of LRUs dramatically to 80, 100, or
128 (avoid 64 & 96 to avoid a bufwaits storm) which may help keep users
from locking each other out of an LRU. You get minimal bufwaits if
LRUs equals the number of concurrent users which is a rare possibility
indeed. The point is if you have 1000 concurrent users bufwaits will
naturally be higher. This check is most significant when watched over
time.
Art S. Kagel