RE: partition buffer summary - Btree percentage
Posted in 1999
Topics: Performance & Tuning, Installation, Setup & Upgrades, Server Administration, Platform-Specific Issues
Hi all. We recently upgraded from 7.23.UC1 to 7.31.UC2, HP-UX 10.20.
Performance was okay for about 3 days then it got bad. onstat -P showed a
high Btree # (95-99%), onstat -R showed a lot of Medium-High's. Both of
these results could point to the 115327 bug described here. We called
Informix and they had us increase our LRUS, CLEANERS, and NUMAIOVPS. We
bounced the database yesterday at noon to put them into effect. Everything
ran good all afternoon and evening until this morning. Our update
statistics scripts ran early this morning.
Does anyone know if this bug 115327 has a cumulative effect? In other
words, I theorize that if I bounce the database now, whatever has
accumulated will be cleared out and we should be good for a while. We will
not run any update statistics scripts for now to keep it as clean as
possible. This is only a temporary solution until Informix gets back to us
with patch information for the bug. Comments, anyone? I would appreciate
the feedback while we struggle through this mess.
(Yes, I am aware of the sql to run and rebuilding the indexes, but that will
take a lot of time so I'm looking at it from a different angle.)
Thanks, Dianne
-----Original Message-----
From: plsmith@my-deja.com [mailto:plsmith@my-deja.com]
Sent: September 21,1999 11:40 PM
To: informix-list@iiug.org
Subject: Re: partition buffer summary - Btree percentage
Doug, good luck with the stress tests - I would be interested to hear
how it goes. The version I have is 7.30.UC7XM and seems ok so far.
I think the interim 7.30 release with the fix is 7.30.UC10 - which isn't
out yet.
Pete
Logica
In article <6uPF3.2929$9W2.5801@news2.mco>,
"Doug Agnew" <dagnew@charlottepipe.com> wrote:
> It is sorta fixed in 7.30UC7 --- we are using version 7.30UC7XK1, and
it
> seems to be okay there (at least our small test system didn't show the
> buffer problem). Will probably be stressing it this week as we
upgrade
> SAP....
>
> Doug
>
> Art S. Kagel wrote in message <37E7BA3B.DC191AA1@bloomberg.net>...
> >I believe it is fixed in 7.31UC3 and 7.30UC7, I relocated my desk and
> >cannot find the note on the subject so I'm going from memory here. I
do
> >not think there is a fix contemplated for 7.2x so upgrades MAY have
to
> >rebuild indexes to eliminate the problem.
> >
> >Art S. Kagel
> >
> >"Carlson@WHSmith" wrote:
> >>
> >> Any idea in which version this bug is fixed?
> >>
> >> John Carlson
> >> Informix DBA
> >> WHSmith USA
> >>
> >> plsmith@my-deja.com wrote:
> >> >
> >> > Vardan
> >> >
> >> > You have almost certainly run into bug no. 115327 and you should
be
> >> > able to request a patch to fix it.
> >> > It is a bug in 7.2x and 7.3x whereby btree pages can be generated
with
> >> > an invalid page type. This page type is then handled incorrectly
by the
> >> > new 7.30 buffer priority management.
> >> > Dropping and rebuilding affected indexes is a reasonable
workaround,
> but
> >> > the problme could re-occur depending on the habits of you
application.
> >> > The invalid page types are generated when all the rows are
deleted from
> >> > a table thus compressing the index down to a single page. When
rows are
> >> > inserted an invalid page type is then copied through all the leaf
> nodes.
> >> > If you look at onstat -R you will see that the majority of pages
are in
> >> > the medium-high category, whereas data pages are medium-low. Leaf
node
> >> > pages should also be medium low, but because of the invalid page
type
> >> > they are placed in the medium-high category, thus squeezing out
the
> data
> >> > pages. If you install a patch you should see a reversal of these
> >> > percentages and an increase in your read cache.
> >> > If you talk to Informix tech support refer them to case No.
861366
> which
> >> > contains a reproducable example.
> >> > I have a small script which will identify invalid pages types
(sql
> >> > courtesy of John Miller, thanks ).
> >> >
> >> > email me if you would like a copy.
> >> >
> >> > hth
> >> > Pete Smith
> >> > Logica
> >> >
> >> > In article <7s5n3g$mq7$1@nnrp1.deja.com>,
> >> > Vardan Aroustamian <vaar@geocities.com> wrote:
> >> > > Hello,
> >> > >
> >> > > I'm new to SAP environment and would like to get some input
> >> > > from experienced users.
> >> > >
> >> > > Anyone knows of some "usual" percentage for Data vs. Btree
> >> > > pages in the buffer pool (from onstat -P output).
> >> > >
> >> > > Currently I have
> >> > >
> >> > > Percentages:
> >> > > Data 20.09
> >> > > Btree 77.75
> >> > > Other 2.16
> >> > >
> >> > > I heard about some bug when Btree is more than 90%.
> >> > > I'm not sure will I reach that number.
> >> > >
> >> > > Informix 7.30.UC7XB
> >> > > SunOS 5.5.1
> >> > >
> >> > > Database ~50GB
> >> > > SAP 4.5
> >> > >
> >> > > Actually I believe this is not SAP specific question.
> >> > > Anyway, thanks for any information.
> >> > >
> >> > > Vardan
> >> > >
> >> > > --
> >> > > Vardan Aroustamian
> >> > > vaar@geocities.com
> >> > >
> >> > > Sent via Deja.com http://www.deja.com/
> >> > > Share what you know. Learn what you don't.
> >> > >
> >> >
> >> > Sent via Deja.com http://www.deja.com/
> >> > Share what you know. Learn what you don't.
>
>
Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.
Dianne,
Yep, it is cumulative. The longer the server is up, the worse it gets
(until it hits 98+%). With SAP, we were bouncing the server about every 12
hours.
We're running 7.30.UC7XK, which allows you to turn off the buffer
prioritization and getting ready to start running UC7XK1, which is SUPPOSED
to fix the problem.
Thanks,
Doug
dianne.pendleton@autodesk.com wrote in message
<7stl0t$6uj$1@news.xmission.com>...
>
>Hi all. We recently upgraded from 7.23.UC1 to 7.31.UC2, HP-UX 10.20.
>Performance was okay for about 3 days then it got bad. onstat -P showed a
>high Btree # (95-99%), onstat -R showed a lot of Medium-High's. Both of
>these results could point to the 115327 bug described here. We called
>Informix and they had us increase our LRUS, CLEANERS, and NUMAIOVPS. We
>bounced the database yesterday at noon to put them into effect. Everything
>ran good all afternoon and evening until this morning. Our update
>statistics scripts ran early this morning.
>Does anyone know if this bug 115327 has a cumulative effect? In other
>words, I theorize that if I bounce the database now, whatever has
>accumulated will be cleared out and we should be good for a while. We will
>not run any update statistics scripts for now to keep it as clean as
>possible. This is only a temporary solution until Informix gets back to us
>with patch information for the bug. Comments, anyone? I would appreciate
>the feedback while we struggle through this mess.
>(Yes, I am aware of the sql to run and rebuilding the indexes, but that
will
>take a lot of time so I'm looking at it from a different angle.)
>
>Thanks, Dianne
>
>-----Original Message-----
>From: plsmith@my-deja.com [mailto:plsmith@my-deja.com]
>Sent: September 21,1999 11:40 PM
>To: informix-list@iiug.org
>Subject: Re: partition buffer summary - Btree percentage
>
>
>Doug, good luck with the stress tests - I would be interested to hear
>how it goes. The version I have is 7.30.UC7XM and seems ok so far.
>
>I think the interim 7.30 release with the fix is 7.30.UC10 - which isn't
>out yet.
>
>Pete
>Logica
>
>
>In article <6uPF3.2929$9W2.5801@news2.mco>,
> "Doug Agnew" <dagnew@charlottepipe.com> wrote:
>> It is sorta fixed in 7.30UC7 --- we are using version 7.30UC7XK1, and
>it
>> seems to be okay there (at least our small test system didn't show the
>> buffer problem). Will probably be stressing it this week as we
>upgrade
>> SAP....
>>
>> Doug
>>
>> Art S. Kagel wrote in message <37E7BA3B.DC191AA1@bloomberg.net>...
>> >I believe it is fixed in 7.31UC3 and 7.30UC7, I relocated my desk and
>> >cannot find the note on the subject so I'm going from memory here. I
>do
>> >not think there is a fix contemplated for 7.2x so upgrades MAY have
>to
>> >rebuild indexes to eliminate the problem.
>> >
>> >Art S. Kagel
>> >
>> >"Carlson@WHSmith" wrote:
>> >>
>> >> Any idea in which version this bug is fixed?
>> >>
>> >> John Carlson
>> >> Informix DBA
>> >> WHSmith USA
>> >>
>> >> plsmith@my-deja.com wrote:
>> >> >
>> >> > Vardan
>> >> >
>> >> > You have almost certainly run into bug no. 115327 and you should
>be
>> >> > able to request a patch to fix it.
>> >> > It is a bug in 7.2x and 7.3x whereby btree pages can be generated
>with
>> >> > an invalid page type. This page type is then handled incorrectly
>by the
>> >> > new 7.30 buffer priority management.
>> >> > Dropping and rebuilding affected indexes is a reasonable
>workaround,
>> but
>> >> > the problme could re-occur depending on the habits of you
>application.
>> >> > The invalid page types are generated when all the rows are
>deleted from
>> >> > a table thus compressing the index down to a single page. When
>rows are
>> >> > inserted an invalid page type is then copied through all the leaf
>> nodes.
>> >> > If you look at onstat -R you will see that the majority of pages
>are in
>> >> > the medium-high category, whereas data pages are medium-low. Leaf
>node
>> >> > pages should also be medium low, but because of the invalid page
>type
>> >> > they are placed in the medium-high category, thus squeezing out
>the
>> data
>> >> > pages. If you install a patch you should see a reversal of these
>> >> > percentages and an increase in your read cache.
>> >> > If you talk to Informix tech support refer them to case No.
>861366
>> which
>> >> > contains a reproducable example.
>> >> > I have a small script which will identify invalid pages types
>(sql
>> >> > courtesy of John Miller, thanks ).
>> >> >
>> >> > email me if you would like a copy.
>> >> >
>> >> > hth
>> >> > Pete Smith
>> >> > Logica
>> >> >
>> >> > In article <7s5n3g$mq7$1@nnrp1.deja.com>,
>> >> > Vardan Aroustamian <vaar@geocities.com> wrote:
>> >> > > Hello,
>> >> > >
>> >> > > I'm new to SAP environment and would like to get some input
>> >> > > from experienced users.
>> >> > >
>> >> > > Anyone knows of some "usual" percentage for Data vs. Btree
>> >> > > pages in the buffer pool (from onstat -P output).
>> >> > >
>> >> > > Currently I have
>> >> > >
>> >> > > Percentages:
>> >> > > Data 20.09
>> >> > > Btree 77.75
>> >> > > Other 2.16
>> >> > >
>> >> > > I heard about some bug when Btree is more than 90%.
>> >> > > I'm not sure will I reach that number.
>> >> > >
>> >> > > Informix 7.30.UC7XB
>> >> > > SunOS 5.5.1
>> >> > >
>> >> > > Database ~50GB
>> >> > > SAP 4.5
>> >> > >
>> >> > > Actually I believe this is not SAP specific question.
>> >> > > Anyway, thanks for any information.
>> >> > >
>> >> > > Vardan
>> >> > >
>> >> > > --
>> >> > > Vardan Aroustamian
>> >> > > vaar@geocities.com
>> >> > >
>> >> > > Sent via Deja.com http://www.deja.com/
>> >> > > Share what you know. Learn what you don't.
>> >> > >
>> >> >
>> >> > Sent via Deja.com http://www.deja.com/
>> >> > Share what you know. Learn what you don't.
>>
>>
>
>
>Sent via Deja.com http://www.deja.com/
>Share what you know. Learn what you don't.