iidbms draining swap space 
Author Message
 iidbms draining swap space

We just moved our primary database from 6.4 on SCO to ingres II on Solaris 7
x86.  The coversion went very well, but it hit the fan later.  The iidbms
process appears to be draining the swap space.  When we run out of
swap...crash.

Top reveals that iimerge, with the pid of the iidbms, just keeps growing,
even at night when the database is mostly idle (it serves a web site).  This
is also confirmed with
 ps -elf | sort -n +9 -10 |tail -22

How can I stop the iidbms from consuming the swap space?

Thanks,
Todd Boewe
Database Engineering
www.lioninc.com



Sun, 24 Feb 2002 03:00:00 GMT
 iidbms draining swap space

Quote:

> We just moved our primary database from 6.4 on SCO to ingres II on Solaris 7
> x86.  The coversion went very well, but it hit the fan later.  The iidbms
> process appears to be draining the swap space.  When we run out of
> swap...crash.

> Top reveals that iimerge, with the pid of the iidbms, just keeps growing,
> even at night when the database is mostly idle (it serves a web site).  This
> is also confirmed with
>  ps -elf | sort -n +9 -10 |tail -22

> How can I stop the iidbms from consuming the swap space?

Hm, is this a supported configuration? I dunno.

I notice that OI2.0 (& probably Ingres II) need a lot more swap than
6.4;
e.g. one of our DBMSs (2.0/9712 on SunOS 5.6 Sparc) runs at around 1GB:

ps -lp 15360
 F S   UID   PID  PPID  C PRI NI     ADDR     SZ    WCHAN TTY      TIME
CMD
 8 S   101 15360     1  0  43 20 6f6f7578 136454 6bc04576 ?       32:41
iimerge
(pagesize 8192)

so I should add a load of swap files in the first place. I haven't heard
of
serious memory leak troubles - would it help to start a sideline DBMS,
and
find out what sort of activity causes this behaviour, i.e. is it a DB
connect/disconnect, update, any query, or whatever, running a ps -lp
<DBMSpid>
to check ?

Richard.
--
Erratic news-swerver! :. post+e-mail to be sure I get your msg.

The Open University is not responsible for content herein, which
may be incorrect and is used at reader's own risk.



Mon, 25 Feb 2002 03:00:00 GMT
 iidbms draining swap space
Hi Todd,

    In short - you cant stop iidbms draining swap space. You can throw more
    swap space at the problem - possibly even dynamically.

    How big is your swap area? I have 2G, typically I would be seeing swap
    only 30% used on most platforms (OI 1.2 on DUNIX). I generally get worried
    about the 50% used mark and start hunting for processes to kill off.

    I find ps -uingres -oUFMT is very usefull in determining memory leakage
    and it can be a good guide to who is using the swap area most.
    To measure swap utilisation I run swapon -s and record the output regurly.

    If you have a genuine memory leakage problem then report this to CA.

    Also check the max-per-proc-data-size setting on your box.
    Try running a '/sbin/sysconfig -q proc' and check the values of that field
    and max-per-proc-address-space. Unfortunatly I'm not familiar enough with
    SCO to say if you have such a utility available

    Martin Bowes

Quote:

> We just moved our primary database from 6.4 on SCO to ingres II on Solaris 7
> x86.  The coversion went very well, but it hit the fan later.  The iidbms
> process appears to be draining the swap space.  When we run out of
> swap...crash.

> Top reveals that iimerge, with the pid of the iidbms, just keeps growing,
> even at night when the database is mostly idle (it serves a web site).  This
> is also confirmed with
>  ps -elf | sort -n +9 -10 |tail -22

> How can I stop the iidbms from consuming the swap space?

> Thanks,
> Todd Boewe
> Database Engineering
> www.lioninc.com

--
Random Duckman Quote #5:
Duckman - You! The son of poor Okey Mountain pigs.  How'd you afford
          med school?
Cornfed - Two words... Vivisection volunteer.  Sure it hurt, but I'm
         told as a result of banging me on the head with a large hammer
         & then slicing open my stomach & digging around in my insides
         they were able to rule out two theories on why fat people
         perspire.
Duckman - Oh.. a worthy cause, so shall we get on with it.
Cornfed - Bend over, grab the desk & grit your teeth.


Mon, 25 Feb 2002 03:00:00 GMT
 iidbms draining swap space


Quote:
> Hi Todd,

>     In short - you cant stop iidbms draining swap space. You can throw
more
>     swap space at the problem - possibly even dynamically.

>     How big is your swap area? I have 2G, typically I would be seeing swap
>     only 30% used on most platforms (OI 1.2 on DUNIX). I generally get
worried
>     about the 50% used mark and start hunting for processes to kill off.

We have just under 2 G.  The iimerge for the iidbms is taking approximately
1 M per minute.

Quote:
>     I find ps -uingres -oUFMT is very usefull in determining memory
leakage
>     and it can be a good guide to who is using the swap area most.
>     To measure swap utilisation I run swapon -s and record the output
regurly.

>     If you have a genuine memory leakage problem then report this to CA.

It has been reported.

Quote:
>     Also check the max-per-proc-data-size setting on your box.

It is unlimited.  What should it be?  Could putting a limit lead to an
earlier crash?

Quote:
>     Try running a '/sbin/sysconfig -q proc' and check the values of that
field
>     and max-per-proc-address-space. Unfortunatly I'm not familiar enough
with
>     SCO to say if you have such a utility available

We are on Intel Solaris 7.  CA now says that the problem is Solaris 7 for
intel which is unsupported and we need to move back to 2.6.  That's the end
of our support.  So, for those of you with more experience, do you recommend
making the move backward (which means buying a new box, unloading, etc. to
keep 24x7 web access) or trying to find a fix to the problem.

Quote:
>     Martin Bowes

> > We just moved our primary database from 6.4 on SCO to ingres II on
Solaris 7
> > x86.  The coversion went very well, but it hit the fan later.  The
iidbms
> > process appears to be draining the swap space.  When we run out of
> > swap...crash.

> > Top reveals that iimerge, with the pid of the iidbms, just keeps
growing,
> > even at night when the database is mostly idle (it serves a web site).
This
> > is also confirmed with
> >  ps -elf | sort -n +9 -10 |tail -22

> > How can I stop the iidbms from consuming the swap space?

> > Thanks,
> > Todd Boewe
> > Database Engineering
> > www.lioninc.com

> --
> Random Duckman Quote #5:
> Duckman - You! The son of poor Okey Mountain pigs.  How'd you afford
>           med school?
> Cornfed - Two words... Vivisection volunteer.  Sure it hurt, but I'm
>          told as a result of banging me on the head with a large hammer
>          & then slicing open my stomach & digging around in my insides
>          they were able to rule out two theories on why fat people
>          perspire.
> Duckman - Oh.. a worthy cause, so shall we get on with it.
> Cornfed - Bend over, grab the desk & grit your teeth.



Tue, 26 Feb 2002 03:00:00 GMT
 iidbms draining swap space
Hi Todd,

    Theres no doubt about it. Thats a memory leakage problem of fairly major
    proportions. Anything else is like the Captain of the Titanic saying
    "There is no need for concern, we have simply stopped to take on a little
    ice".

    Setting memory process limits would only make the crash happen sooner
    rather than later. At the rate you've indicated you can only just survive
    a day as it is. Which for a 24x7 shop is about as usefull as an ashtray on
    a motorcycle.

    I suspect your only option is to bring up 2.6.

    Martin Bowes

Quote:



> > Hi Todd,

> >     In short - you cant stop iidbms draining swap space. You can throw
> more
> >     swap space at the problem - possibly even dynamically.

> >     How big is your swap area? I have 2G, typically I would be seeing swap
> >     only 30% used on most platforms (OI 1.2 on DUNIX). I generally get
> worried
> >     about the 50% used mark and start hunting for processes to kill off.

> We have just under 2 G.  The iimerge for the iidbms is taking approximately
> 1 M per minute.

> >     I find ps -uingres -oUFMT is very usefull in determining memory
> leakage
> >     and it can be a good guide to who is using the swap area most.
> >     To measure swap utilisation I run swapon -s and record the output
> regurly.

> >     If you have a genuine memory leakage problem then report this to CA.

> It has been reported.

> >     Also check the max-per-proc-data-size setting on your box.

> It is unlimited.  What should it be?  Could putting a limit lead to an
> earlier crash?

> >     Try running a '/sbin/sysconfig -q proc' and check the values of that
> field
> >     and max-per-proc-address-space. Unfortunatly I'm not familiar enough
> with
> >     SCO to say if you have such a utility available

> We are on Intel Solaris 7.  CA now says that the problem is Solaris 7 for
> intel which is unsupported and we need to move back to 2.6.  That's the end
> of our support.  So, for those of you with more experience, do you recommend
> making the move backward (which means buying a new box, unloading, etc. to
> keep 24x7 web access) or trying to find a fix to the problem.

> >     Martin Bowes

> > > We just moved our primary database from 6.4 on SCO to ingres II on
> Solaris 7
> > > x86.  The coversion went very well, but it hit the fan later.  The
> iidbms
> > > process appears to be draining the swap space.  When we run out of
> > > swap...crash.

> > > Top reveals that iimerge, with the pid of the iidbms, just keeps
> growing,
> > > even at night when the database is mostly idle (it serves a web site).
> This
> > > is also confirmed with
> > >  ps -elf | sort -n +9 -10 |tail -22

> > > How can I stop the iidbms from consuming the swap space?

> > > Thanks,
> > > Todd Boewe
> > > Database Engineering
> > > www.lioninc.com

> > --
> > Random Duckman Quote #5:
> > Duckman - You! The son of poor Okey Mountain pigs.  How'd you afford
> >           med school?
> > Cornfed - Two words... Vivisection volunteer.  Sure it hurt, but I'm
> >          told as a result of banging me on the head with a large hammer
> >          & then slicing open my stomach & digging around in my insides
> >          they were able to rule out two theories on why fat people
> >          perspire.
> > Duckman - Oh.. a worthy cause, so shall we get on with it.
> > Cornfed - Bend over, grab the desk & grit your teeth.

--
Random Earthworm Jim Quote #18:
Bob(The Killer GoldFish) -
    If you want something done right, hire a guy with a monkey for a head.
    Thats what I always say.


Thu, 28 Feb 2002 03:00:00 GMT
 
 [ 5 post ] 

 Relevant Pages 

1. Lost in space/space lost, a space management problem

2. Oracle 8.x/Solaris Swap Space Configuration

3. Swap Space Requirements with 9i RAC?

4. Swap space on Solaris

5. Oracle 7.2.3 swap space misread on Solaris 2.6

6. dbsnmp with huge heap causes low swap space

7. Error:no swap space to grow stack (solaris)

8. swap space declines on ultrasparc1/Solaris2.5 with Oracle Workgroup Server

9. dbsnmp with huge heap causes low swap space

10. Merge join exhausting swap space

11. Error 3026 Not enough space, but 2G space

12. unallocated space and unused space


 
Powered by phpBB® Forum Software