Thread: Limited performance on multi core server

Limited performance on multi core server

From
Matthew Lunnon
Date:
Hi,

I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
postgres 7.4.3.  This has been recompiled on the server for 64 stored
procedure parameters, (I assume this makes postgres 64 bit but are not
sure).  When the server gets under load from database connections
executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
limit at about 30-35% usage with no iowait reported. If I run a simple
select at this time it takes 5 seconds, the same query runs in 300
millis when the server is not under load so it seems that the database
is not performing well even though there is plenty of spare CPU.  There
does not appear to be large amounts of disk IO and my database is about
5.5G so this should fit comfortably in RAM.

changes to postgresql.sql:

max_connections = 500
shared_buffers = 96000
sort_mem = 10240
effective_cache_size = 1000000

Does anyone have any ideas what my bottle neck might be and what I can
do about it?

Thanks for any help.

Matthew.

Re: Limited performance on multi core server

From
"Steinar H. Gunderson"
Date:
On Wed, Dec 12, 2007 at 10:16:43AM +0000, Matthew Lunnon wrote:
> Does anyone have any ideas what my bottle neck might be and what I can do
> about it?

Your bottleneck is that you are using a very old version of PostgreSQL. Try
8.2 or (if you can) the 8.3 beta series -- it scales a _lot_ better in this
kind of situation.

/* Steinar */
--
Homepage: http://www.sesse.net/

Re: Limited performance on multi core server

From
"Claus Guttesen"
Date:
> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
> postgres 7.4.3.  This has been recompiled on the server for 64 stored
> procedure parameters, (I assume this makes postgres 64 bit but are not
> sure).  When the server gets under load from database connections
> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
> limit at about 30-35% usage with no iowait reported. If I run a simple
> select at this time it takes 5 seconds, the same query runs in 300
> millis when the server is not under load so it seems that the database
> is not performing well even though there is plenty of spare CPU.  There
> does not appear to be large amounts of disk IO and my database is about
> 5.5G so this should fit comfortably in RAM.
>
> changes to postgresql.sql:
>
> max_connections = 500
> shared_buffers = 96000
> sort_mem = 10240
> effective_cache_size = 1000000
>
> Does anyone have any ideas what my bottle neck might be and what I can
> do about it?

You might want to lower shared_buffers to a lower value. Mine is set
at 32768. Is your db performing complex sort? Remember that this value
is per connection. Maby 1024. effective_cache_size should also be
lowered to something like 32768. As far as I understand shared_buffers
and effective_cache_size have to be altered "in reverse", ie. when
lowering one the other can be raised.

HTH.

--
regards
Claus

When lenity and cruelty play for a kingdom,
the gentlest gamester is the soonest winner.

Shakespeare

Re: Limited performance on multi core server

From
"Claus Guttesen"
Date:
> > Does anyone have any ideas what my bottle neck might be and what I can do
> > about it?
>
> Your bottleneck is that you are using a very old version of PostgreSQL. Try
> 8.2 or (if you can) the 8.3 beta series -- it scales a _lot_ better in this
> kind of situation.

You won't know until you've seen what queries are performed. Changing
db-parameters is a short-term solution, upgrading to a newer version
does require some planning.

--
regards
Claus

When lenity and cruelty play for a kingdom,
the gentlest gamester is the soonest winner.

Shakespeare

Re: Limited performance on multi core server

From
Sven Geisler
Date:
Hi Matthew,

I know exactly what you experience.
We had a 4-way DC Opteron and Pg 7.4 too.
You should monitor context switches.


First suggest upgrade to 8.2.5 because the scale up is much better with 8.2.

You need to limit the number of concurrent queries to less than 8 (8
cores) if you need to stay with Pg 7.4.

The memory setting is looking good to me. I would increase sort_mem and
effective_cache_size, but this would solve your problem.

Best regards
Sven.



Matthew Lunnon schrieb:
> Hi,
>
> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
> postgres 7.4.3.  This has been recompiled on the server for 64 stored
> procedure parameters, (I assume this makes postgres 64 bit but are not
> sure).  When the server gets under load from database connections
> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
> limit at about 30-35% usage with no iowait reported. If I run a simple
> select at this time it takes 5 seconds, the same query runs in 300
> millis when the server is not under load so it seems that the database
> is not performing well even though there is plenty of spare CPU.  There
> does not appear to be large amounts of disk IO and my database is about
> 5.5G so this should fit comfortably in RAM.
>
> changes to postgresql.sql:
>
> max_connections = 500
> shared_buffers = 96000
> sort_mem = 10240
> effective_cache_size = 1000000
>
> Does anyone have any ideas what my bottle neck might be and what I can
> do about it?
>
> Thanks for any help.
>
> Matthew.
>
> ---------------------------(end of broadcast)---------------------------
> TIP 6: explain analyze is your friend

--
Sven Geisler <sgeisler@aeccom.com>   Tel +49.30.921017.81  Fax .50
Senior Developer, AEC/communications GmbH & Co. KG Berlin, Germany

Re: Limited performance on multi core server

From
Matthew Lunnon
Date:
Limiting the queries was our initial thought but we then hit a problem with connection pooling which didn't implement a fifo algorithm.  Looks like I'll have to look deeper into the connection pooling.

So you think the problem might be context switching on the server, I'll take a closer look at the this

Thanks

Matthew

Sven Geisler wrote:
Hi Matthew,

I know exactly what you experience.
We had a 4-way DC Opteron and Pg 7.4 too.
You should monitor context switches.


First suggest upgrade to 8.2.5 because the scale up is much better with 8.2.

You need to limit the number of concurrent queries to less than 8 (8
cores) if you need to stay with Pg 7.4.

The memory setting is looking good to me. I would increase sort_mem and
effective_cache_size, but this would solve your problem.

Best regards
Sven.



Matthew Lunnon schrieb: 
Hi,

I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
postgres 7.4.3.  This has been recompiled on the server for 64 stored
procedure parameters, (I assume this makes postgres 64 bit but are not
sure).  When the server gets under load from database connections
executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
limit at about 30-35% usage with no iowait reported. If I run a simple
select at this time it takes 5 seconds, the same query runs in 300
millis when the server is not under load so it seems that the database
is not performing well even though there is plenty of spare CPU.  There
does not appear to be large amounts of disk IO and my database is about
5.5G so this should fit comfortably in RAM.

changes to postgresql.sql:

max_connections = 500
shared_buffers = 96000
sort_mem = 10240
effective_cache_size = 1000000

Does anyone have any ideas what my bottle neck might be and what I can
do about it?

Thanks for any help.

Matthew.

---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend   
 

-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--

Re: Limited performance on multi core server

From
Matthew Lunnon
Date:
Ah I was afraid of that.  Maybe I'll have to come out of the dark ages.

Matthew

Steinar H. Gunderson wrote:
On Wed, Dec 12, 2007 at 10:16:43AM +0000, Matthew Lunnon wrote: 
Does anyone have any ideas what my bottle neck might be and what I can do 
about it?   
Your bottleneck is that you are using a very old version of PostgreSQL. Try
8.2 or (if you can) the 8.3 beta series -- it scales a _lot_ better in this
kind of situation.

/* Steinar */ 

-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--

Re: Limited performance on multi core server

From
Matthew Lunnon
Date:
Thanks for the information Claus,  Why would reducing the effective cache size help the processor usage?  It seems that there is plenty of resources on the box although I can see that 10MB of sort space could mount up if we had 500 connections but at the moment we do not have anything like that number.

Thanks
Matthew.

Claus Guttesen wrote:
I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
postgres 7.4.3.  This has been recompiled on the server for 64 stored
procedure parameters, (I assume this makes postgres 64 bit but are not
sure).  When the server gets under load from database connections
executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
limit at about 30-35% usage with no iowait reported. If I run a simple
select at this time it takes 5 seconds, the same query runs in 300
millis when the server is not under load so it seems that the database
is not performing well even though there is plenty of spare CPU.  There
does not appear to be large amounts of disk IO and my database is about
5.5G so this should fit comfortably in RAM.

changes to postgresql.sql:

max_connections = 500
shared_buffers = 96000
sort_mem = 10240
effective_cache_size = 1000000

Does anyone have any ideas what my bottle neck might be and what I can
do about it?   
You might want to lower shared_buffers to a lower value. Mine is set
at 32768. Is your db performing complex sort? Remember that this value
is per connection. Maby 1024. effective_cache_size should also be
lowered to something like 32768. As far as I understand shared_buffers
and effective_cache_size have to be altered "in reverse", ie. when
lowering one the other can be raised.

HTH.
 

-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--

Re: Limited performance on multi core server

From
"Claus Guttesen"
Date:
>  Thanks for the information Claus,  Why would reducing the effective cache
> size help the processor usage?  It seems that there is plenty of resources
> on the box although I can see that 10MB of sort space could mount up if we
> had 500 connections but at the moment we do not have anything like that
> number.

There is a discussion at
http://archives.postgresql.org/pgsql-performance/2005-06/msg00477.php
which can give a clearer picture. But in general rasing values can be
counterproductive.

If you know that you won't need more than 250 connections that would
be a reasonable value. You may wan't to read
http://www.varlena.com/GeneralBits/Tidbits/annotated_conf_e.html as
well. This has some rules of thumb on the settings for 7.4.x.

--
regards
Claus

When lenity and cruelty play for a kingdom,
the gentlest gamester is the soonest winner.

Shakespeare

Re: Limited performance on multi core server

From
Greg Smith
Date:
On Wed, 12 Dec 2007, Matthew Lunnon wrote:

> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
> postgres 7.4.3.
> shared_buffers = 96000

As you've already been told repeatedly 7.4 is a release from long before
optimizations to work well on a multi-core server like this.  I'll only
add that because of those problems, larger values of shared_buffers were
sometimes counter-productive with these old versions.  You should try
reducing that to the 10,000-50000 range and see if things improve; that's
the general range that was effective with 7.4.  Continue to set
effective_cache_size to a large value so that the optimizer knows how much
RAM is really available.

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD

Re: Limited performance on multi core server

From
Sven Geisler
Date:
Hi Matthew,

The context switching isn't the issue. This is an indicator which is
useful to identify your problem.

What kind of application do you running? Can you limit the database clients?

We have a web application based on apache running. We have a limit
number of apache processes which are able to connect the database.
We use that to reduce the number of concurrent queries.
The apache does the rest for us - the apache does queue incoming http
request if all workers are busy. The configuration helps us to solve the
 performance issue with to much concurrent queries.

I assume that you already checked you application and each sql query is
necessary and tuned as best as you can.

Regards
Sven.

Matthew Lunnon schrieb:
> Limiting the queries was our initial thought but we then hit a problem
> with connection pooling which didn't implement a fifo algorithm.  Looks
> like I'll have to look deeper into the connection pooling.
>
> So you think the problem might be context switching on the server, I'll
> take a closer look at the this
>
> Thanks
>
> Matthew
>
> Sven Geisler wrote:
>> Hi Matthew,
>>
>> I know exactly what you experience.
>> We had a 4-way DC Opteron and Pg 7.4 too.
>> You should monitor context switches.
>>
>>
>> First suggest upgrade to 8.2.5 because the scale up is much better with 8.2.
>>
>> You need to limit the number of concurrent queries to less than 8 (8
>> cores) if you need to stay with Pg 7.4.
>>
>> The memory setting is looking good to me. I would increase sort_mem and
>> effective_cache_size, but this would solve your problem.
>>
>> Best regards
>> Sven.
>>
>>
>>
>> Matthew Lunnon schrieb:
>>
>>> Hi,
>>>
>>> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
>>> postgres 7.4.3.  This has been recompiled on the server for 64 stored
>>> procedure parameters, (I assume this makes postgres 64 bit but are not
>>> sure).  When the server gets under load from database connections
>>> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
>>> limit at about 30-35% usage with no iowait reported. If I run a simple
>>> select at this time it takes 5 seconds, the same query runs in 300
>>> millis when the server is not under load so it seems that the database
>>> is not performing well even though there is plenty of spare CPU.  There
>>> does not appear to be large amounts of disk IO and my database is about
>>> 5.5G so this should fit comfortably in RAM.
>>>
>>> changes to postgresql.sql:
>>>
>>> max_connections = 500
>>> shared_buffers = 96000
>>> sort_mem = 10240
>>> effective_cache_size = 1000000
>>>
>>> Does anyone have any ideas what my bottle neck might be and what I can
>>> do about it?
>>>
>>> Thanks for any help.
>>>
>>> Matthew.
>>>
>>> ---------------------------(end of broadcast)---------------------------
>>> TIP 6: explain analyze is your friend
>>>
>>
>>
>
> --
> Matthew Lunnon
> Technical Consultant
> RWA Ltd.
>
>  mlunnon@rwa-net.co.uk
>  Tel: +44 (0)29 2081 5056
>  www.rwa-net.co.uk
> --
>

--
Sven Geisler <sgeisler@aeccom.com>   Tel +49.30.921017.81  Fax .50
Senior Developer, AEC/communications GmbH & Co. KG Berlin, Germany

Re: Limited performance on multi core server

From
Sven Geisler
Date:
Hi Matthew,

I remember that I also an issue with AMD Opterons before Pg 8.1
There is a specific Opteron behaviour on shared memory locks which adds
a extra "penalty" during the execution time for Pg code before 8.1.
I can you provide my patch for Pg 8.0 which should be adaptable for Pg
7.4 if you can compile PostgreSQL.

But if you can upgrade you should upgrade to Pg 8.2.5 64-bit. The scale
up for your concurrent queries will be great.

Sven.

Matthew Lunnon schrieb:
> Hi,
>
> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
> postgres 7.4.3.  This has been recompiled on the server for 64 stored
> procedure parameters, (I assume this makes postgres 64 bit but are not
> sure).  When the server gets under load from database connections
> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
> limit at about 30-35% usage with no iowait reported. If I run a simple
> select at this time it takes 5 seconds, the same query runs in 300
> millis when the server is not under load so it seems that the database
> is not performing well even though there is plenty of spare CPU.  There
> does not appear to be large amounts of disk IO and my database is about
> 5.5G so this should fit comfortably in RAM.
>
> changes to postgresql.sql:
>
> max_connections = 500
> shared_buffers = 96000
> sort_mem = 10240
> effective_cache_size = 1000000
>
> Does anyone have any ideas what my bottle neck might be and what I can
> do about it?
>
> Thanks for any help.
>
> Matthew.
>
> ---------------------------(end of broadcast)---------------------------
> TIP 6: explain analyze is your friend

--
Sven Geisler <sgeisler@aeccom.com>   Tel +49.30.921017.81  Fax .50
Senior Developer, AEC/communications GmbH & Co. KG Berlin, Germany

Re: Limited performance on multi core server

From
Matthew Lunnon
Date:
Hi Sven,

yes the patch would be great if you could send it to me, we have already had to compile postgres to up the number of function parameters from 32 to 64.

Meanwhile I will try and persuade my colleagues to consider the upgrade option.

Thanks
Matthew

Sven Geisler wrote:
Hi Matthew,

I remember that I also an issue with AMD Opterons before Pg 8.1
There is a specific Opteron behaviour on shared memory locks which adds
a extra "penalty" during the execution time for Pg code before 8.1.
I can you provide my patch for Pg 8.0 which should be adaptable for Pg
7.4 if you can compile PostgreSQL.

But if you can upgrade you should upgrade to Pg 8.2.5 64-bit. The scale
up for your concurrent queries will be great.

Sven.

Matthew Lunnon schrieb: 
Hi,

I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
postgres 7.4.3.  This has been recompiled on the server for 64 stored
procedure parameters, (I assume this makes postgres 64 bit but are not
sure).  When the server gets under load from database connections
executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
limit at about 30-35% usage with no iowait reported. If I run a simple
select at this time it takes 5 seconds, the same query runs in 300
millis when the server is not under load so it seems that the database
is not performing well even though there is plenty of spare CPU.  There
does not appear to be large amounts of disk IO and my database is about
5.5G so this should fit comfortably in RAM.

changes to postgresql.sql:

max_connections = 500
shared_buffers = 96000
sort_mem = 10240
effective_cache_size = 1000000

Does anyone have any ideas what my bottle neck might be and what I can
do about it?

Thanks for any help.

Matthew.

---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend   
 

-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--

Re: Limited performance on multi core server

From
Matthew Lunnon
Date:
Hi Sven,

Yes I have done a reasonable amount of query tuning.  The application is a web service using an apache/resin combination at the front end, we have thought about using resin threads to limit the number of connections but are worried about backing up connections in apache and getting some overflow here.  But some kind of limiting of connections is probably required.

Thanks
Matthew

Sven Geisler wrote:
Hi Matthew,

The context switching isn't the issue. This is an indicator which is
useful to identify your problem.

What kind of application do you running? Can you limit the database clients?

We have a web application based on apache running. We have a limit
number of apache processes which are able to connect the database.
We use that to reduce the number of concurrent queries.
The apache does the rest for us - the apache does queue incoming http
request if all workers are busy. The configuration helps us to solve theperformance issue with to much concurrent queries.

I assume that you already checked you application and each sql query is
necessary and tuned as best as you can.

Regards
Sven.

Matthew Lunnon schrieb: 
Limiting the queries was our initial thought but we then hit a problem
with connection pooling which didn't implement a fifo algorithm.  Looks
like I'll have to look deeper into the connection pooling.

So you think the problem might be context switching on the server, I'll
take a closer look at the this

Thanks

Matthew

Sven Geisler wrote:   
Hi Matthew,

I know exactly what you experience.
We had a 4-way DC Opteron and Pg 7.4 too.
You should monitor context switches.


First suggest upgrade to 8.2.5 because the scale up is much better with 8.2.

You need to limit the number of concurrent queries to less than 8 (8
cores) if you need to stay with Pg 7.4.

The memory setting is looking good to me. I would increase sort_mem and
effective_cache_size, but this would solve your problem.

Best regards
Sven.



Matthew Lunnon schrieb:      
Hi,

I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
postgres 7.4.3.  This has been recompiled on the server for 64 stored
procedure parameters, (I assume this makes postgres 64 bit but are not
sure).  When the server gets under load from database connections
executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
limit at about 30-35% usage with no iowait reported. If I run a simple
select at this time it takes 5 seconds, the same query runs in 300
millis when the server is not under load so it seems that the database
is not performing well even though there is plenty of spare CPU.  There
does not appear to be large amounts of disk IO and my database is about
5.5G so this should fit comfortably in RAM.

changes to postgresql.sql:

max_connections = 500
shared_buffers = 96000
sort_mem = 10240
effective_cache_size = 1000000

Does anyone have any ideas what my bottle neck might be and what I can
do about it?

Thanks for any help.

Matthew.

---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend          
       
-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--
   
 

-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--

Re: Limited performance on multi core server

From
Sven Geisler
Date:
Hi Matthew,

Please not that is no official patch, but it works with our Opteron
server without any problems. You should run the regression test after
you adapt it for Pg 7.4.

Sven.

Matthew Lunnon schrieb:
> Hi Sven,
>
> yes the patch would be great if you could send it to me, we have already
> had to compile postgres to up the number of function parameters from 32
> to 64.
>
> Meanwhile I will try and persuade my colleagues to consider the upgrade
> option.
>
> Thanks
> Matthew
>
> Sven Geisler wrote:
>> Hi Matthew,
>>
>> I remember that I also an issue with AMD Opterons before Pg 8.1
>> There is a specific Opteron behaviour on shared memory locks which adds
>> a extra "penalty" during the execution time for Pg code before 8.1.
>> I can you provide my patch for Pg 8.0 which should be adaptable for Pg
>> 7.4 if you can compile PostgreSQL.
>>
>> But if you can upgrade you should upgrade to Pg 8.2.5 64-bit. The scale
>> up for your concurrent queries will be great.
>>
>> Sven.
>>
>> Matthew Lunnon schrieb:
>>
>>> Hi,
>>>
>>> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
>>> postgres 7.4.3.  This has been recompiled on the server for 64 stored
>>> procedure parameters, (I assume this makes postgres 64 bit but are not
>>> sure).  When the server gets under load from database connections
>>> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
>>> limit at about 30-35% usage with no iowait reported. If I run a simple
>>> select at this time it takes 5 seconds, the same query runs in 300
>>> millis when the server is not under load so it seems that the database
>>> is not performing well even though there is plenty of spare CPU.  There
>>> does not appear to be large amounts of disk IO and my database is about
>>> 5.5G so this should fit comfortably in RAM.
>>>
>>> changes to postgresql.sql:
>>>
>>> max_connections = 500
>>> shared_buffers = 96000
>>> sort_mem = 10240
>>> effective_cache_size = 1000000
>>>
>>> Does anyone have any ideas what my bottle neck might be and what I can
>>> do about it?
>>>
>>> Thanks for any help.
>>>
>>> Matthew.
>>>
>>> ---------------------------(end of broadcast)---------------------------
>>> TIP 6: explain analyze is your friend
>>>
>>
>>
>
> --
> Matthew Lunnon
> Technical Consultant
> RWA Ltd.
>
>  mlunnon@rwa-net.co.uk
>  Tel: +44 (0)29 2081 5056
>  www.rwa-net.co.uk
> --
>

--
Sven Geisler <sgeisler@aeccom.com>   Tel +49.30.921017.81  Fax .50
Senior Developer, AEC/communications GmbH & Co. KG Berlin, Germany
*** postgresql-8.0.3/src/include/storage/s_lock.h.orig    Sun Aug 28 20:41:44 2005
--- postgresql-8.0.3/src/include/storage/s_lock.h    Fri Sep  9 14:58:44 2005
***************
*** 120,132 ****
  {
      register slock_t _res = 1;

!     /* Use a non-locking test before asserting the bus lock */
      __asm__ __volatile__(
-         "    cmpb    $0,%1    \n"
-         "    jne        1f        \n"
-         "    lock            \n"
          "    xchgb    %0,%1    \n"
-         "1: \n"
  :        "+q"(_res), "+m"(*lock)
  :
  :        "memory", "cc");
--- 120,128 ----
  {
      register slock_t _res = 1;

!     /* xchg implies a LOCK prefix, so no need to say LOCK explicitly */
      __asm__ __volatile__(
          "    xchgb    %0,%1    \n"
  :        "+q"(_res), "+m"(*lock)
  :
  :        "memory", "cc");
*** postgresql-8.0.3/src/backend/storage/lmgr/s_lock.c.orig    Fri Aug 26 10:47:35 2005
--- postgresql-8.0.3/src/backend/storage/lmgr/s_lock.c    Sat Sep 10 19:19:46 2005
***************
*** 17,22 ****
--- 17,23 ----

  #include <time.h>
  #include <unistd.h>
+ #include <sched.h>

  #include "storage/s_lock.h"
  #include "miscadmin.h"
***************
*** 83,89 ****

      int            spins = 0;
      int            delays = 0;
!     int            cur_delay = MIN_DELAY_CSEC;

      while (TAS(lock))
      {
--- 84,90 ----

      int            spins = 0;
      int            delays = 0;
!     int            cur_delay = 0;

      while (TAS(lock))
      {
***************
*** 96,102 ****
              if (++delays > NUM_DELAYS)
                  s_lock_stuck(lock, file, line);

!             pg_usleep(cur_delay * 10000L);

  #if defined(S_LOCK_TEST)
              fprintf(stdout, "*");
--- 97,110 ----
              if (++delays > NUM_DELAYS)
                  s_lock_stuck(lock, file, line);

!             if (cur_delay == 0)
!             {
!               /* first time through, try just a sched_yield */
!               sched_yield();
!               cur_delay = MIN_DELAY_CSEC;
!             }
!             else
!               pg_usleep(cur_delay * 10000L);

  #if defined(S_LOCK_TEST)
              fprintf(stdout, "*");

Re: Limited performance on multi core server

From
Sven Geisler
Date:
Hi Matthew,

The apache is able to queue 1024 request. Reducing the number of
max_clients was the key to deal with the problem of to much concurrent
queries. We have monitored less concurrent http request after we
decrease max_clients.

We also have introduce a global statement timeout to stop long running
queries.

Both together protect our database server. The problem we had was only
to find the values for our application.

Sven.

Matthew Lunnon schrieb:
> Hi Sven,
>
> Yes I have done a reasonable amount of query tuning.  The application is
> a web service using an apache/resin combination at the front end, we
> have thought about using resin threads to limit the number of
> connections but are worried about backing up connections in apache and
> getting some overflow here.  But some kind of limiting of connections is
> probably required.
>
> Thanks
> Matthew
>
> Sven Geisler wrote:
>> Hi Matthew,
>>
>> The context switching isn't the issue. This is an indicator which is
>> useful to identify your problem.
>>
>> What kind of application do you running? Can you limit the database clients?
>>
>> We have a web application based on apache running. We have a limit
>> number of apache processes which are able to connect the database.
>> We use that to reduce the number of concurrent queries.
>> The apache does the rest for us - the apache does queue incoming http
>> request if all workers are busy. The configuration helps us to solve the
>>  performance issue with to much concurrent queries.
>>
>> I assume that you already checked you application and each sql query is
>> necessary and tuned as best as you can.
>>
>> Regards
>> Sven.
>>
>> Matthew Lunnon schrieb:
>>
>>> Limiting the queries was our initial thought but we then hit a problem
>>> with connection pooling which didn't implement a fifo algorithm.  Looks
>>> like I'll have to look deeper into the connection pooling.
>>>
>>> So you think the problem might be context switching on the server, I'll
>>> take a closer look at the this
>>>
>>> Thanks
>>>
>>> Matthew
>>>
>>> Sven Geisler wrote:
>>>
>>>> Hi Matthew,
>>>>
>>>> I know exactly what you experience.
>>>> We had a 4-way DC Opteron and Pg 7.4 too.
>>>> You should monitor context switches.
>>>>
>>>>
>>>> First suggest upgrade to 8.2.5 because the scale up is much better with 8.2.
>>>>
>>>> You need to limit the number of concurrent queries to less than 8 (8
>>>> cores) if you need to stay with Pg 7.4.
>>>>
>>>> The memory setting is looking good to me. I would increase sort_mem and
>>>> effective_cache_size, but this would solve your problem.
>>>>
>>>> Best regards
>>>> Sven.
>>>>
>>>>
>>>>
>>>> Matthew Lunnon schrieb:
>>>>
>>>>
>>>>> Hi,
>>>>>
>>>>> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
>>>>> postgres 7.4.3.  This has been recompiled on the server for 64 stored
>>>>> procedure parameters, (I assume this makes postgres 64 bit but are not
>>>>> sure).  When the server gets under load from database connections
>>>>> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
>>>>> limit at about 30-35% usage with no iowait reported. If I run a simple
>>>>> select at this time it takes 5 seconds, the same query runs in 300
>>>>> millis when the server is not under load so it seems that the database
>>>>> is not performing well even though there is plenty of spare CPU.  There
>>>>> does not appear to be large amounts of disk IO and my database is about
>>>>> 5.5G so this should fit comfortably in RAM.
>>>>>
>>>>> changes to postgresql.sql:
>>>>>
>>>>> max_connections = 500
>>>>> shared_buffers = 96000
>>>>> sort_mem = 10240
>>>>> effective_cache_size = 1000000
>>>>>
>>>>> Does anyone have any ideas what my bottle neck might be and what I can
>>>>> do about it?
>>>>>
>>>>> Thanks for any help.
>>>>>
>>>>> Matthew.
>>>>>
>>>>> ---------------------------(end of broadcast)---------------------------
>>>>> TIP 6: explain analyze is your friend
>>>>>
>>>>>
>>>>
>>>>
>>> --
>>> Matthew Lunnon
>>> Technical Consultant
>>> RWA Ltd.
>>>
>>>  mlunnon@rwa-net.co.uk
>>>  Tel: +44 (0)29 2081 5056
>>>  www.rwa-net.co.uk
>>> --
>>>
>>>
>>
>>
>
> --
> Matthew Lunnon
> Technical Consultant
> RWA Ltd.
>
>  mlunnon@rwa-net.co.uk
>  Tel: +44 (0)29 2081 5056
>  www.rwa-net.co.uk
> --
>

--
Sven Geisler <sgeisler@aeccom.com>   Tel +49.30.921017.81  Fax .50
Senior Developer, AEC/communications GmbH & Co. KG Berlin, Germany

Re: Limited performance on multi core server

From
Matthew Lunnon
Date:
Hi Sven,

Does this mean that one option I have is to use a multi core Intel based server instead of an AMD based server?

Matthew

Sven Geisler wrote:
Hi Matthew,

I remember that I also an issue with AMD Opterons before Pg 8.1
There is a specific Opteron behaviour on shared memory locks which adds
a extra "penalty" during the execution time for Pg code before 8.1.
I can you provide my patch for Pg 8.0 which should be adaptable for Pg
7.4 if you can compile PostgreSQL.

But if you can upgrade you should upgrade to Pg 8.2.5 64-bit. The scale
up for your concurrent queries will be great.

Sven.

Matthew Lunnon schrieb: 
Hi,

I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
postgres 7.4.3.  This has been recompiled on the server for 64 stored
procedure parameters, (I assume this makes postgres 64 bit but are not
sure).  When the server gets under load from database connections
executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
limit at about 30-35% usage with no iowait reported. If I run a simple
select at this time it takes 5 seconds, the same query runs in 300
millis when the server is not under load so it seems that the database
is not performing well even though there is plenty of spare CPU.  There
does not appear to be large amounts of disk IO and my database is about
5.5G so this should fit comfortably in RAM.

changes to postgresql.sql:

max_connections = 500
shared_buffers = 96000
sort_mem = 10240
effective_cache_size = 1000000

Does anyone have any ideas what my bottle neck might be and what I can
do about it?

Thanks for any help.

Matthew.

---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend   
 

-- 
Matthew Lunnon
Technical Consultant
RWA Ltd.
mlunnon@rwa-net.co.ukTel: +44 (0)29 2081 5056www.rwa-net.co.uk
--

Re: Limited performance on multi core server

From
Sven Geisler
Date:
Hi Matthew,

You should be upgrade to Pg 8.2.5.
We did test on different hardware. The bigest step is to use Pg 8.2.5.
The high number of context switches which we saw before simple disappeared.


In regards to you question about XEONs: You will have the similar issue
with a XEON box.
We tried different boxes (4-way DC, 8-way, 8-way DC and 4-way QC).
The improvement compared with a 4-way SC wan't really better under Pg
8.0. I'm not saying that a 4-way SC is faster than the 4-way QC in terms
of handling more concurrent queries. But there was never a huge step as
you would expect from the spend money.

Regards
Sven.

Matthew Lunnon schrieb:
> Hi Sven,
>
> Does this mean that one option I have is to use a multi core Intel based
> server instead of an AMD based server?
>
> Matthew
>
> Sven Geisler wrote:
>> Hi Matthew,
>>
>> I remember that I also an issue with AMD Opterons before Pg 8.1
>> There is a specific Opteron behaviour on shared memory locks which adds
>> a extra "penalty" during the execution time for Pg code before 8.1.
>> I can you provide my patch for Pg 8.0 which should be adaptable for Pg
>> 7.4 if you can compile PostgreSQL.
>>
>> But if you can upgrade you should upgrade to Pg 8.2.5 64-bit. The scale
>> up for your concurrent queries will be great.
>>
>> Sven.
>>
>> Matthew Lunnon schrieb:
>>
>>> Hi,
>>>
>>> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
>>> postgres 7.4.3.  This has been recompiled on the server for 64 stored
>>> procedure parameters, (I assume this makes postgres 64 bit but are not
>>> sure).  When the server gets under load from database connections
>>> executing reads, lets say 20 - 40 concurrent reads, the CPU's seem to
>>> limit at about 30-35% usage with no iowait reported. If I run a simple
>>> select at this time it takes 5 seconds, the same query runs in 300
>>> millis when the server is not under load so it seems that the database
>>> is not performing well even though there is plenty of spare CPU.  There
>>> does not appear to be large amounts of disk IO and my database is about
>>> 5.5G so this should fit comfortably in RAM.
>>>
>>> changes to postgresql.sql:
>>>
>>> max_connections = 500
>>> shared_buffers = 96000
>>> sort_mem = 10240
>>> effective_cache_size = 1000000
>>>
>>> Does anyone have any ideas what my bottle neck might be and what I can
>>> do about it?
>>>
>>> Thanks for any help.
>>>
>>> Matthew.
>>>
>>> ---------------------------(end of broadcast)---------------------------
>>> TIP 6: explain analyze is your friend
>>>
>>
>>
>
> --
> Matthew Lunnon
> Technical Consultant
> RWA Ltd.
>
>  mlunnon@rwa-net.co.uk
>  Tel: +44 (0)29 2081 5056
>  www.rwa-net.co.uk
> --
>

--
Sven Geisler <sgeisler@aeccom.com>   Tel +49.30.921017.81  Fax .50
Senior Developer, AEC/communications GmbH & Co. KG Berlin, Germany

Re: Limited performance on multi core server

From
"Joshua D. Drake"
Date:
Matthew Lunnon wrote:
> Ah I was afraid of that.  Maybe I'll have to come out of the dark ages.

Yes :) but ignore the comment about the 8.3 Beta series. It is Beta for
a reason, that means testing only, no production.

Sincerely,

Joshua D. Drake

>> Your bottleneck is that you are using a very old version of PostgreSQL. Try
>> 8.2 or (if you can) the 8.3 beta series -- it scales a _lot_ better in this
>> kind of situation.
>>
>> /* Steinar */


Re: Limited performance on multi core server

From
"Joshua D. Drake"
Date:
Matthew Lunnon wrote:
> Hi Sven,
>
> Does this mean that one option I have is to use a multi core Intel based
> server instead of an AMD based server?

Wow... hold on.

>> I remember that I also an issue with AMD Opterons before Pg 8.1
>> There is a specific Opteron behaviour on shared memory locks which adds
>> a extra "penalty" during the execution time for Pg code before 8.1.

I would like to see some proof of this.


>> I can you provide my patch for Pg 8.0 which should be adaptable for Pg
>> 7.4 if you can compile PostgreSQL.
>>

The real answer here is to upgrade to 8.1.10 or 8.2.5. 64bit if you can.

>>> I have a 4 * dual core 64bit AMD OPTERON server with 16G of RAM, running
>>> postgres 7.4.3.  This has been recompiled on the server for 64 stored
>>> procedure parameters, (I assume this makes postgres 64 bit but are not

This type of machine, assuming you have decent IO available will
generally be very fast with anything >= 8.1.

>>> changes to postgresql.sql:
>>>
>>> max_connections = 500
>>> shared_buffers = 96000

Way to high for 7.4. Way to low for 8.1 or above (based on what you
mention your work load is)

>>> sort_mem = 10240

Based on your specs this actually may be fine but I would suggest
reviewing it after you upgrade.

>>> effective_cache_size = 1000000
>>>

Again, too low for 8.1 or above.

Sincerely,

Joshua D. Drake

--
The PostgreSQL Company: http://www.commandprompt.com/


Re: Limited performance on multi core server

From
Simon Riggs
Date:
On Wed, 2007-12-12 at 07:38 -0800, Joshua D. Drake wrote:
> Matthew Lunnon wrote:
> > Ah I was afraid of that.  Maybe I'll have to come out of the dark ages.
>
> Yes :) but ignore the comment about the 8.3 Beta series. It is Beta for
> a reason, that means testing only, no production.

Matthew,

Some benchmark comparisons would be really useful though, so please
don't be dissuaded from looking at 8.3. It's planned to be released
after Christmas, so we're probably around 2 business weeks away from
release if you ignore the holiday period.

Everything you've said suggests you've hit the scalability limit of 7.4,
which had a buffer manager that got worse with larger settings, fixed in
8.0. Most of the scalability stuff has been added since then and 8.3
looks to be really fast, but we would still like some more performance
numbers.

--
  Simon Riggs
  2ndQuadrant  http://www.2ndQuadrant.com


Re: Limited performance on multi core server

From
"Joshua D. Drake"
Date:
Simon Riggs wrote:
> On Wed, 2007-12-12 at 07:38 -0800, Joshua D. Drake wrote:
>> Matthew Lunnon wrote:
>>> Ah I was afraid of that.  Maybe I'll have to come out of the dark ages.
>> Yes :) but ignore the comment about the 8.3 Beta series. It is Beta for
>> a reason, that means testing only, no production.
>
> Matthew,
>
> Some benchmark comparisons would be really useful though, so please
> don't be dissuaded from looking at 8.3.

Right, I did say "testing" which is always good.

Sincerely,

Joshua D. Drake


Re: Limited performance on multi core server

From
Heikki Linnakangas
Date:
Matthew Lunnon wrote:
> Ah I was afraid of that.  Maybe I'll have to come out of the dark ages.

At the very least, upgrade to latest 7.4 minor version. It probably
won't help with you're performance, but 7.4.3 is very old. There's been
a *lot* of bug fixes between 7.4.3 and 7.4.18, including fixes for
security vulnerabilities and data corruption bugs.

> Steinar H. Gunderson wrote:
>> On Wed, Dec 12, 2007 at 10:16:43AM +0000, Matthew Lunnon wrote:
>>
>>> Does anyone have any ideas what my bottle neck might be and what I
>>> can do about it?
>>>
>>
>> Your bottleneck is that you are using a very old version of
>> PostgreSQL. Try
>> 8.2 or (if you can) the 8.3 beta series -- it scales a _lot_ better in
>> this
>> kind of situation.
>>
>> /* Steinar */
>>
>


--
   Heikki Linnakangas
   EnterpriseDB   http://www.enterprisedb.com