Thread: Linux 2.6 kernel, tuned for use with databases - does that apply to Postgresql too?

Linux 2.6 kernel, tuned for use with databases - does that apply to Postgresql too?

From
Martin_Hurst@dom.com
Date:
I was reading the following Information Week article:
http://www.informationweek.com/story/showArticle.jhtml?articleID=10100565&pgno=2

------------------------------------------------
Linux is likely to make inroads in the enterprise with the release later
this year of the 2.6 kernel, tuned for use with databases, Witham says.
Changes to the Linux 2.6 kernel will let programs access more data with
greater reliability and run heavier processing loads. It will improve the
performance of databases on eight- and 16-way symmetric multiprocessor
servers, include new file systems that provide faster access to data with
greater reliability, and make it easier to manage and configure storage on
large arrays.
------------------------------------------------

Are there plans to have Postgresql take advantage of the new Linux 2.6
kernel?
-Martin



Re: Linux 2.6 kernel,

From
Arjen van der Meijden
Date:
Afaik, most of such changes to the new kernel branch are independent of
whatever you run on top of it.
There are improvements in scheduling of processes and threads, there are
improvements on diskacessing algortims (I think) and such.

All databases have more or less similar requirements, like: Fast random
disk access, efficient multithreading/multiprocessing
With such kernelimprovements, postgresql will probably automatically
take advantage of such improvements.

On the other hand, there might be a few kernelparameters that need
tuning and those should be explained in the performance-tuning
documents, as soon as there is more information on their behaviour and
such :)

Regards,

Arjen

> Martin_Hurst@dom.com wrote:
>
> I was reading the following Information Week article:
> http://www.informationweek.com/story/showArticle.jhtml?article
ID=10100565&pgno=2

------------------------------------------------
Linux is likely to make inroads in the enterprise with the release later
this year of the 2.6 kernel, tuned for use with databases, Witham says.
Changes to the Linux 2.6 kernel will let programs access more data with
greater reliability and run heavier processing loads. It will improve
the performance of databases on eight- and 16-way symmetric
multiprocessor servers, include new file systems that provide faster
access to data with greater reliability, and make it easier to manage
and configure storage on large arrays.
------------------------------------------------

Are there plans to have Postgresql take advantage of the new Linux 2.6
kernel? -Martin



---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

http://archives.postgresql.org



Re: Linux 2.6 kernel, tuned for use with databases -

From
Ron Johnson
Date:
On Fri, 2003-06-06 at 08:21, Martin_Hurst@dom.com wrote:
> I was reading the following Information Week article:
> http://www.informationweek.com/story/showArticle.jhtml?articleID=10100565&pgno=2
>
> ------------------------------------------------
> Linux is likely to make inroads in the enterprise with the release later
> this year of the 2.6 kernel, tuned for use with databases, Witham says.
> Changes to the Linux 2.6 kernel will let programs access more data with
> greater reliability and run heavier processing loads. It will improve the
> performance of databases on eight- and 16-way symmetric multiprocessor
> servers, include new file systems that provide faster access to data with
> greater reliability, and make it easier to manage and configure storage on
> large arrays.
> ------------------------------------------------
>
> Are there plans to have Postgresql take advantage of the new Linux 2.6
> kernel?

Since PostgreSQL is a Unix RDBMS, not a Linux DBMS, and already
uses multiple CPUs if possible, does it really matter how Linux
changes?

Shouldn't PG automatically take advantage of new features, maybe
after a backup/restore to new disk drives/filesystems?

--
+-----------------------------------------------------------+
| Ron Johnson, Jr.     Home: ron.l.johnson@cox.net          |
| Jefferson, LA  USA   http://members.cox.net/ron.l.johnson |
|                                                           |
| Regarding war zones: "There's nothing sacrosanct about a  |
| hotel with a bunch of journalists in it."                 |
|     Marine Lt. Gen. Bernard E. Trainor (Retired)          |
+-----------------------------------------------------------+


Re: Linux 2.6 kernel,

From
Alvaro Herrera
Date:
On Sun, Jun 08, 2003 at 09:55:19PM +0200, Arjen van der Meijden wrote:
> Afaik, most of such changes to the new kernel branch are independent of
> whatever you run on top of it.
> There are improvements in scheduling of processes and threads, there are
> improvements on diskacessing algortims (I think) and such.

However there are a couple of things that probably will need some
effort from the PgSQL camp.  Examples are the futexes thing (fast
mostly-user-space mutexes that are said to be much faster than SysV
semaphores) and async I/O.

--
Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
Y dijo Dios: "Que sea Satanás, para que la gente no me culpe de todo a mí."
"Y que hayan abogados, para que la gente no culpe de todo a Satanás"

Re: Linux 2.6 kernel,

From
Bruce Momjian
Date:
Alvaro Herrera wrote:
> On Sun, Jun 08, 2003 at 09:55:19PM +0200, Arjen van der Meijden wrote:
> > Afaik, most of such changes to the new kernel branch are independent of
> > whatever you run on top of it.
> > There are improvements in scheduling of processes and threads, there are
> > improvements on diskacessing algortims (I think) and such.
>
> However there are a couple of things that probably will need some
> effort from the PgSQL camp.  Examples are the futexes thing (fast
> mostly-user-space mutexes that are said to be much faster than SysV
> semaphores) and async I/O.

Of course, we use SysV semaphores only to sleep on while waiting for an
event, like a lock release, so there isn't much win there --- we already
have user-space mutexes in shared memory on all platforms.

And on async, I don't think we have a sure benefit from that either.

--
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Re: Linux 2.6 kernel, tuned for use with databases -

From
Alvaro Herrera
Date:
On Sun, Jun 08, 2003 at 10:10:04PM -0500, Ron Johnson wrote:
> On Fri, 2003-06-06 at 08:21, Martin_Hurst@dom.com wrote:
> > ------------------------------------------------
> > Linux is likely to make inroads in the enterprise with the release later
> > this year of the 2.6 kernel, tuned for use with databases, Witham says.
>
> Since PostgreSQL is a Unix RDBMS, not a Linux DBMS, and already
> uses multiple CPUs if possible, does it really matter how Linux
> changes?

Yes it does matter.  The Linux port can be improved and more performance
can be obtained.  This doesn't mean the other ports are going to be
abandoned.

OTOH the quoted article contained more hype and marketing buzz than real
substance AFAICT...

--
Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
"Acepta los honores y aplausos y perderas tu libertad"

Re: Linux 2.6 kernel, tuned for use with databases - does

From
"scott.marlowe"
Date:
On Fri, 6 Jun 2003 Martin_Hurst@dom.com wrote:

> I was reading the following Information Week article:
> http://www.informationweek.com/story/showArticle.jhtml?articleID=10100565&pgno=2
>
> ------------------------------------------------
> Linux is likely to make inroads in the enterprise with the release later
> this year of the 2.6 kernel, tuned for use with databases, Witham says.
> Changes to the Linux 2.6 kernel will let programs access more data with
> greater reliability and run heavier processing loads. It will improve the
> performance of databases on eight- and 16-way symmetric multiprocessor
> servers, include new file systems that provide faster access to data with
> greater reliability, and make it easier to manage and configure storage on
> large arrays.
> ------------------------------------------------
>
> Are there plans to have Postgresql take advantage of the new Linux 2.6
> kernel?

Nothing really needs to be done.  If Linux 2.6 is faster at SMP, then
postgresql gets to be faster too.  The APIs haven't changed, the kernel's
just been made to run more efficiently under high load, so Postgresql will
just run faster.

That's what's so nice about good modular design.  The pgsql team don't
have to pay a lot of attention to the differences between bsd, linux,
hpux, solaris, etc... to get good performance.

I rememeber seeing noticeable performance gains going from the 2.2 kernel
to the 2.4 kernel, I'm sure I'll see them again when 2.6 goes live.