Re: 答复:About "Context-switch storm"problem

--=-xze2+Hj3kU8pEvj/aKr/
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

1- Of course, we could. Problem appeared in both 8.1.8 and 8.1.15 after
an initial upgrade.
Today database is not online due to a migration to another technology.
It's not our database (a customer one) so we have no access to it at
this moment.=20

2- One of the major on-line shops of Spain. This only happened under
heavy load condition (that's under Christmas campaign), with a lot
queries per minutes, some of then heavy search on database using tsearch
modules. Under low load, there was no problem.

After upgrading to 8.2 and splitting our database in two (one database
only for catalog searches, another database for standard operation)
problem disappeared.

Our context switches where about 5000-10000 per second or less under low
load scaling up to 300.000 per second under heavy load. Most of CPU time
was spent on context switches and not on operating DB.
The database was 20 gigabytes size over a Quad Intel Xeon-DP (HT
enabled) at 2.6Ghz with 16GB RAM memory running on a Red Hat 3.X with
kernel 2.4.

First we upgraded to Red Hat 5 in order to get a kernel 2.6, then we
installed postgres 8.2 and finally we splitted database.
Typical load decreased from 80-100% on high load periods to 5% or less
per server with these changes. Context switches drop to 5000 or less per
second.

I'm think that splitting database wouldn't have been needed, but anyway
the changes were made to deal with higher loads in future... although
six months later the shop was migrated to SQL Server + ASP.NET
technology (decision was taken one year before and was not due to these
problems :-) ).

The problem was due to several factors: kernel 2.4 + postgresql 8.2 +
Intel Xeon DP =20

-----Original Message-----
From: 黄永å=AB
To: 'Iñigo Martinez Lasala'
Cc: 'pgsql-admin'
Subject: 答复: About "Context-switch storm" problem
Date: Mon, 31 May 2010 10:43:55 +0800

Dear Iñigo Martinez Lasalaï¼=8C

Thank you for your replay and advice.=20

My company is a software application vendor, and our customer is useing
PostgreSQL 8.1.3.

When our application is runing , the DB sluggish suddenly and
application stop.

To prove that it is not the problem of the application ,We need some
evidence to certified this version have problem and show to our
customer.

And Can I ask you several questions?

=20

1ã€=81 Can you recreate the "Context-switch storm" in the your old
version PostgreSQL 8.1.15 ?

2ã€=81 If you can ,could you please tell me your test case ï¼=9F

=20

Thank you so muchï¼=81

=20

Best regards,

Ray Huang

=20

=20
____________________________________________________________ ____________
发件äº=BA: Iñigo Martinez Lasala [mailto:imartinez=
@vectorsf.com]=20
发送时间: 2010å¹=B45æœ=8828æ—=A5 =
15:43
收件äº=BA: 黄永å=AB
抄送: pgsql-admin
主题: Re: About "Context-switch storm" problem


=20

Hi, Yongwei

We finally upgraded our database server to last PostgreSQL 8.2
available.

Since 8.2 does not include any major change (8.3 does indeed a lot of
them) we had no problem with this upgrade. And believe, our database was
really complex. About 600 tables with 300 stored procedures, wide use of
embedded functions from different packages (including tsearch2) and lot
of triggers and views. Not problems at all.=20

Migrating to 8.3 or 8.4 is another history. So, my recommendation is to
migrate to 8.2.17

-----Original Message-----
From: 黄永å=AB
To: imartinez@vectorsf.com
Subject: About "Context-switch storm" problem
Date: Fri, 28 May 2010 13:31:54 +0800

Helloï¼=8CIñigo Martinez Lasala !

=20

I read your mail about â€=9CContext-switch storm in 8.1.15â€=9D f=
rom postgres
website.

I think I meet the same problem as you now .

Recently my companyâ€=99s postgres DB server sluggish suddenly with =
a hight
Context-switching value as below:

=20

2010-04-07 04:03:15 procs memory swap
io system cpu

2010-04-07 04:03:15 r b swpd free buff cache si so bi
bo in cs us sy id wa

2010-04-07 14:04:27 3 0 0 2361272 272684 3096148 0 0 3
1445 973 14230 7 8 84 0

2010-04-07 14:05:27 2 0 0 2361092 272684 3096220 0 0 3
1804 1029 31852 8 10 81 1

2010-04-07 14:06:27 1 0 0 2362236 272684 3096564 0 0 3
1865 1135 19689 9 9 81 0

2010-04-07 14:07:27 1 0 0 2348400 272720 3101836 0 0 3
1582 1182 149461 15 17 67 0

2010-04-07 14:08:27 3 0 0 2392028 272840 3107600 0 0 3
3093 1275 203196 24 23 53 1

2010-04-07 14:09:27 3 1 0 2386224 272916 3107960 0 0 3
2486 1331 193299 26 22 52 0

2010-04-07 14:10:27 34 0 0 2332320 272980 3107944 0 0 3
1692 1082 214309 24 22 54 0

2010-04-07 14:11:27 1 0 0 2407432 273028 3108092 0 0 6
2770 1540 76643 29 13 57 1

2010-04-07 14:12:27 9 0 0 2358968 273104 3108388 0 0 7
2639 1466 10603 22 6 72 1

=20

=20

I

May I know your solution after that ?

=20

=20

=20

Thank you !

Best wishes,

Ray Huang







--=-xze2+Hj3kU8pEvj/aKr/
Content-Type: text/html; charset="utf-8"
Content-Transfer-Encoding: 7bit








1- Of course, we could. Problem appeared in both 8.1.8 and 8.1.15 after an initial upgrade.

Today database is not online due to a migration to another technology. It's not our database (a customer one) so we have no access to it at this moment.



2- One of the major on-line shops of Spain. This only happened under heavy load condition (that's under Christmas campaign), with a lot queries per minutes, some of then heavy search on database using tsearch modules. Under low load, there was no problem.



After upgrading to 8.2 and splitting our database in two (one database only for catalog searches, another database for standard operation) problem disappeared.



Our context switches where about 5000-10000 per second or less under low load scaling up to 300.000 per second under heavy load. Most of CPU time was spent on context switches and not on operating DB.

The database was 20 gigabytes size over a Quad Intel Xeon-DP (HT enabled) at 2.6Ghz with 16GB RAM memory running on a Red Hat 3.X with kernel 2.4.



First we upgraded to Red Hat 5 in order to get a kernel 2.6, then we installed postgres 8.2 and finally we splitted database.

Typical load decreased from 80-100% on high load periods to 5% or less per server with these changes. Context switches drop to 5000 or less per second.



I'm think that splitting database wouldn't have been needed, but anyway the changes were made to deal with higher loads in future... although six months later the shop was migrated to SQL Server + ASP.NET technology (decision was taken one year before and was not due to these problems :-) ).



The problem was due to several factors: kernel 2.4 + postgresql 8.2 + Intel Xeon DP 



-----Original Message-----

From: 黄永卫 <>

To: 'Iñigo Martinez Lasala' <>

Cc: 'pgsql-admin' <>

Subject: 答复: About "Context-switch storm" problem

Date: Mon, 31 May 2010 10:43:55 +0800



Dear Iñigo Martinez Lasala，



    Thank you for your replay and advice.



My company is a software application vendor, and our customer  is useing PostgreSQL 8.1.3.



When our application is runing , the DB sluggish suddenly and application stop.



To  that it is not the problem of the application ,We need some evidence to certified this version have problem and show to our customer.



    And Can  I ask you    questions?



 



1、  Can you recreate the  "Context-switch storm" in the  your old version PostgreSQL 8.1.15 ?



2、  If you can ,could you please tell me your test case ？



 



Thank you so much！



 



Best regards,



Ray Huang



 

答复: 答复: About "Context-switc

This is a multi-part message in MIME format.

------=_NextPart_000_0075_01CB00DC.1111B790
Content-Type: text/plain;
charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Dear Iñigo Martinez Lasalaï¼=8C =20

=20

Thank you for you information !

=20

=20

_____ =20

发件äº=BA: Iñigo Martinez Lasala =
[mailto:imartinez@vectorsf.com]=20
发送时间: 2010å¹=B45æœ=8831æ—=A5 =
15:52
收件äº=BA: 黄永å=AB
抄送: 'pgsql-admin'
主题: Re: 答复: About "Context-switch storm" =
problem

=20

1- Of course, we could. Problem appeared in both 8.1.8 and 8.1.15 after =
an initial upgrade.
Today database is not online due to a migration to another technology. =
It's not our database (a customer one) so we have no access to it at =
this moment.=20

2- One of the major on-line shops of Spain. This only happened under =
heavy load condition (that's under Christmas campaign), with a lot =
queries per minutes, some of then heavy search on database using tsearch =
modules. Under low load, there was no problem.

After upgrading to 8.2 and splitting our database in two (one database =
only for catalog searches, another database for standard operation) =
problem disappeared.

Our context switches where about 5000-10000 per second or less under low =
load scaling up to 300.000 per second under heavy load. Most of CPU time =
was spent on context switches and not on operating DB.
The database was 20 gigabytes size over a Quad Intel Xeon-DP (HT =
enabled) at 2.6Ghz with 16GB RAM memory running on a Red Hat 3.X with =
kernel 2.4.

First we upgraded to Red Hat 5 in order to get a kernel 2.6, then we =
installed postgres 8.2 and finally we splitted database.
Typical load decreased from 80-100% on high load periods to 5% or less =
per server with these changes. Context switches drop to 5000 or less per =
second.

I'm think that splitting database wouldn't have been needed, but anyway =
the changes were made to deal with higher loads in future... although =
six months later the shop was migrated to SQL Server + ASP.NET =
technology (decision was taken one year before and was not due to these =
problems :-) ).

The problem was due to several factors: kernel 2.4 + postgresql 8.2 + =
Intel Xeon DP =20

-----Original Message-----
From: 黄永å=AB f%3d%20%3cyongwei.huang@gmail.com%3e> >
To: 'Iñigo Martinez Lasala' 20%3cimartinez@vectorsf.com%3e> >
Cc: 'pgsql-admin' >
Subject: 答复: About "Context-switch storm" problem
Date: Mon, 31 May 2010 10:43:55 +0800

Dear Iñigo Martinez Lasalaï¼=8C

Thank you for your replay and advice.=20

My company is a software application vendor, and our customer is useing =
PostgreSQL 8.1.3.

When our application is runing , the DB sluggish suddenly and =
application stop.

To prove that it is not the problem of the application =
,We need some evidence to certified this version have problem and show =
to our customer.

And Can I ask you several questions?

=20

1ã€=81 Can you recreate the "Context-switch storm" in the your =
old version PostgreSQL 8.1.15 ?

2ã€=81 If you can ,could you please tell me your test case =
ï¼=9F

=20

Thank you so muchï¼=81

=20

Best regards,

Ray Huang

=20

_____ =20

发件äº=BA: Iñigo Martinez Lasala =
[mailto:imartinez@vectorsf.com]=20
发送时间: 2010å¹=B45æœ=8828æ—=A5 =
15:43
收件äº=BA: 黄永å=AB
抄送: pgsql-admin
主题: Re: About "Context-switch storm" problem


=20

Hi, Yongwei

We finally upgraded our database server to last PostgreSQL 8.2 =
available.

Since 8.2 does not include any major change (8.3 does indeed a lot of =
them) we had no problem with this upgrade. And believe, our database was =
really complex. About 600 tables with 300 stored procedures, wide use of =
embedded functions from different packages (including tsearch2) and lot =
of triggers and views. Not problems at all.=20

Migrating to 8.3 or 8.4 is another history. So, my recommendation is to =
migrate to 8.2.17

-----Original Message-----
From: 黄永å=AB f%3d%20%3cyongwei.huang@gmail.com%3e> >
To: imartinez@vectorsf.com
Subject: About "Context-switch storm" problem
Date: Fri, 28 May 2010 13:31:54 +0800

Helloï¼=8CIñigo Martinez Lasala !

=20

I read your mail about â€=9CContext-switch storm in 8.1.15â€=9D =
from postgres website.

I think I meet the same problem as you now .

Recently my companyâ€=99s postgres DB server sluggish suddenly with =
a hight Context-switching value as below:

=20

2010-04-07 04:03:15 procs memory swap =
io system cpu

2010-04-07 04:03:15 r b swpd free buff cache si so bi =
bo in cs us sy id wa

2010-04-07 14:04:27 3 0 0 2361272 272684 3096148 0 0 3 =
1445 973 14230 7 8 84 0

2010-04-07 14:05:27 2 0 0 2361092 272684 3096220 0 0 3 =
1804 1029 31852 8 10 81 1

2010-04-07 14:06:27 1 0 0 2362236 272684 3096564 0 0 3 =
1865 1135 19689 9 9 81 0

2010-04-07 14:07:27 1 0 0 2348400 272720 3101836 0 0 3 =
1582 1182 149461 15 17 67 0

2010-04-07 14:08:27 3 0 0 2392028 272840 3107600 0 0 3 =
3093 1275 203196 24 23 53 1

2010-04-07 14:09:27 3 1 0 2386224 272916 3107960 0 0 3 =
2486 1331 193299 26 22 52 0

2010-04-07 14:10:27 34 0 0 2332320 272980 3107944 0 0 3 =
1692 1082 214309 24 22 54 0

2010-04-07 14:11:27 1 0 0 2407432 273028 3108092 0 0 6 =
2770 1540 76643 29 13 57 1

2010-04-07 14:12:27 9 0 0 2358968 273104 3108388 0 0 7 =
2639 1466 10603 22 6 72 1

=20

=20

I

May I know your solution after that ?

=20

=20

=20

Thank you !

Best wishes,

Ray Huang








------=_NextPart_000_0075_01CB00DC.1111B790
Content-Type: text/html;
charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

xmlns:o=3D"urn:schemas-microsoft-com:office:office" =
xmlns:w=3D"urn:schemas-microsoft-com:office:word" =
xmlns:st1=3D"urn:schemas-microsoft-com:office:smarttags" =
xmlns=3D"http://www.w3.org/TR/REC-html40">




namespaceuri=3D"urn:schemas-microsoft-com:office:smarttags" =
name=3D"country-region"/>
namespaceuri=3D"urn:schemas-microsoft-com:office:smarttags"
name=3D"place"/>
namespaceuri=3D"urn:schemas-microsoft-com:office:smarttags"
name=3D"chsdate"/>
namespaceuri=3D"urn:schemas-microsoft-com:office:smarttags"
name=3D"chmetcnv"/>