Re: [HACKERS] PostgreSQL on 64 bit Linux

2006-08-21 Thread Alexander Kirpa
> WRT 64-bit and Postgres, it depends on the CPU as to whether you
> see a simple performance benefit.  On the Opteron you will see a
> benefit when doing CPU bound work.  When doing the CPU portion, the
> additional registers of the Opteron running in 64-bit mode are used
> by the compiler to produce a 20-30% boost in performance.  On the
> Xeon in 64-bit mode, the same regions of execution will slow down
> by about 5%.

> Postgres benefits automatically from the larger memory addressing
> of the 64-bit kernel by using the larger I/O cache of Linux.

Main benefit Postgres in 64-bit mode possible only in case dedicated 
DB server on system with RAM > 3GB and use most part of RAM for 
shared buffers and avoid persistent moving buffers between OS cache 
and shared memory. On system with RAM below 2-3GB to difficult found 
serious gain of performance.

Best regards,
 Alexander Kirpa


---(end of broadcast)---
TIP 3: Have you checked our extensive FAQ?

   http://www.postgresql.org/docs/faq


Re: [HACKERS] PostgreSQL on 64 bit Linux

2006-08-22 Thread Alexander Kirpa
On 21 Aug 2006, at 10:48, [EMAIL PROTECTED] wrote:

> On Mon, Aug 21, 2006 at 12:00:58PM +0300, Alexander Kirpa wrote:
> > > WRT 64-bit and Postgres, it depends on the CPU as to whether you
> > > see a simple performance benefit.  On the Opteron you will see a
> > > benefit when doing CPU bound work.  When doing the CPU portion,
> > > the additional registers of the Opteron running in 64-bit mode are
> > > used by the compiler to produce a 20-30% boost in performance.  On
> > > the Xeon in 64-bit mode, the same regions of execution will slow
> > > down by about 5%.
> > 
> > > Postgres benefits automatically from the larger memory addressing
> > > of the 64-bit kernel by using the larger I/O cache of Linux.
> > 
> > Main benefit Postgres in 64-bit mode possible only in case dedicated
> > DB server on system with RAM > 3GB and use most part of RAM for
> > shared buffers and avoid persistent moving buffers between OS cache
> > and shared memory. On system with RAM below 2-3GB to difficult found
> > serious gain of performance.
> 
> This is the main difference between PostgreSQL today - designed for
> 32-bit - when recompiled with a 64-bit compiler.
> 
> The additional registers are barely enough to counter the increased
> cost of processing in 64-bits.
> 
> Cheers,
> mark
Current 32-bit Postgres architecture allow use main benefit
of 64-bit OS - huge memory size for shared buffers.
At current time possible use 2G x 8KB = 16TB as shared memory
and regarding this issue need use (O_DIRECT) to avoid OS cache 
especially in case databases fit to shared memory.

Best regards,
 Alexander Kirpa



---(end of broadcast)---
TIP 9: In versions below 8.0, the planner will ignore your desire to
   choose an index scan if your joining column's datatypes do not
   match


Re: [HACKERS] Spinlocks, yet again: analysis and proposed patches

2005-09-20 Thread Alexander Kirpa
The fact that cmpb isn't helping proves that getting the cache line 
in a
read-only fashion does *not* do enough to protect the xchgb in this 
way.
But maybe another locking instruction would.  Comments?

regards, tom lane

Hi, Tom!

Possible you help next link
'Replay: Unknown Features of the NetBurst Core'
http://www.xbitlabs.com/articles/cpu/display/replay.html

Best regards,
 Alexander Kirpa


---(end of broadcast)---
TIP 1: if posting/reading through Usenet, please send an appropriate
   subscribe-nomail command to [EMAIL PROTECTED] so that your
   message can get through to the mailing list cleanly