After a weekend of pretty heavy usuage (150-200 queries per sec) I can say the
thread thrashing has been fixed with the latest binary.
Our load averages however, are still 2-3x what they were with the older binary.
Even when mysql isn't being queried at all (when apache is shut down) the load
the Perl program.
Perform query.
sleep 10;
perform query.
sleep 10;
Hope this helps,
~Kelly W. Black
-Original Message-
From: heath boutwell [mailto:heathboutwell;yahoo.com]
Sent: Tuesday, November 05, 2002 7:19 AM
To: [EMAIL PROTECTED]
Subject: RE: Thread Thrashing and 3.23.53a
After
Hi.
On Tue 2002-10-29 at 16:01:00 -0800, [EMAIL PROTECTED] wrote:
On Tue, Oct 29, 2002 at 03:56:04PM -0800, heath boutwell wrote:
[...]
Load average of 1.5 seems high when we are only averaging 14 queries
per sec (nothing else going crazy running on the box)
procs
MYTOP reports key eff at 98.30%
+--+---+
| Variable_name| Value |
+--+---+
| Aborted_clients | 58|
| Aborted_connects | 6 |
| Bytes_received | 281234326 |
|
Administrator Asarian-host.org
---
If you were supposed to understand it,
we wouldn't call it code. - FedEx
- Original Message -
From: heath boutwell [EMAIL PROTECTED]
To: [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Sent: Tuesday, October 29, 2002 11:34 PM
Subject: Re: Thread Thrashing
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
On Wednesday 30 October 2002 06:06, Jeremy Zawodny wrote:
On Wed, Oct 30, 2002 at 05:56:52AM +0100, Mark wrote:
Is this all true? Bummer. :( I just upgraded to 3.23.53 (from
.49). I cannot afford to get these huge load averages on my news
On Wed, Oct 30, 2002 at 11:17:44AM +0100, Benjamin Pflugmann wrote:
Hi.
Allow me to disagree. Although most often the symptoms you cite are a
sign that the operation is disk-bound, they are more a sign of
blocking per se, of which the most common cause is disk operation. In
the above case
Heath, I once had similar symptoms when my DNS resolving had problems
and all new connections slept for some seconds while they were
waiting for an response. My work-around was to use --skip-name-resolve
until the DNS problem was solved.
Benjamin,
Thanks for the suggestion but the connections
On Wed, Oct 30, 2002 at 10:20:05AM -0800, heath boutwell wrote:
Heath, I once had similar symptoms when my DNS resolving had problems
and all new connections slept for some seconds while they were
waiting for an response. My work-around was to use --skip-name-resolve
until the DNS problem was
Hi.
On Wed 2002-10-30 at 10:20:05 -0800, [EMAIL PROTECTED] wrote:
Heath, I once had similar symptoms when my DNS resolving had problems
and all new connections slept for some seconds while they were
waiting for an response. My work-around was to use --skip-name-resolve
until the DNS problem
On Tue, Oct 29, 2002 at 01:28:01PM -0800, heath boutwell wrote:
Can anyone that was experiencing the thread thrashing problems with
the other binaries confirm that this has been with resolved with
53a?
I've read others who've said it fixed the problem for them. Did you
check the archives?
--
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
I've read others who've said it fixed the problem
for them. Did you
check the archives?
Jeremy,
Thanks for mytop. Its a great tool and its been a
great help in diagnosing our problems.
I saw where a few of the others claimed the latest
binaries
On Tue, Oct 29, 2002 at 02:34:44PM -0800, heath boutwell wrote:
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
I've read others who've said it fixed the problem
for them. Did you
check the archives?
Jeremy,
Thanks for mytop. Its a great tool and its been a great help in
diagnosing our
On Tue, Oct 29, 2002 at 03:56:04PM -0800, heath boutwell wrote:
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
Well, load average can be misleading.
Can you maybe share the output of vmstat 1 or vmstat 5 for a
minute's worth of time? It's be good to see what the CPUs states look
like.
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
You're disk-bound, not CPU bound. Notice all the idle CPU time? And
the fact that the b column always has one process listed, while r
only does once in a while.
2 gigs of ram -- don't really know why we'd be disk bound. The my.cnf is
pretty well
On Tue, Oct 29, 2002 at 07:37:09PM -0800, heath boutwell wrote:
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
You're disk-bound, not CPU bound. Notice all the idle CPU time? And
the fact that the b column always has one process listed, while r
only does once in a while.
2 gigs of ram --
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
Can you post the output of SHOW VARIABLES?
|
+-+
| back_log| 50
On Tue, Oct 29, 2002 at 08:25:00PM -0800, heath boutwell wrote:
--- Jeremy Zawodny [EMAIL PROTECTED] wrote:
Can you post the output of SHOW VARIABLES?
|
| key_buffer_size | 402649088
PROTECTED]
Sent: Tuesday, October 29, 2002 11:34 PM
Subject: Re: Thread Thrashing and 3.23.53a
Jeremy,
Thanks for mytop. Its a great tool and its been a
great help in diagnosing our problems.
I saw where a few of the others claimed the latest
binaries resolved the issues. Its not clear
On Wed, Oct 30, 2002 at 05:56:52AM +0100, Mark wrote:
Is this all true? Bummer. :( I just upgraded to 3.23.53 (from
.49). I cannot afford to get these huge load averages on my news
server. Nor can I afford thread thrashing. I just had to let myself
be talked into upgrading. See if I still
20 matches
Mail list logo