Thread: Re: [PATCHES] SRA Win32 sync() code

Re: [PATCHES] SRA Win32 sync() code

From
"Merlin Moncure"
Date:
Bruce Momjian wrote:
> So, by my logic, if we have 100 backends all doing updates, we will
need
> 10 * 100 or 1000 writer processes or threads to keep up with that
load.
> That seems quite excessive to me from a context switching and process
> overhead perspective.

Quick point:
A single process using multiple threads dedicated to writing is an
excellent optimization target on the win32 platform, (and if it is
similarly useful on other platforms, so much the better).  To my way of
thinking, this is an ideal approach in the long run.

Multiple processes scheduling writes (even it is only 10), IMO, is a bad
idea because of the way process management on win32 works for various
reasons.

Merlin


Re: [PATCHES] SRA Win32 sync() code

From
Bruce Momjian
Date:
Merlin Moncure wrote:
> Bruce Momjian wrote:
> > So, by my logic, if we have 100 backends all doing updates, we will
> need
> > 10 * 100 or 1000 writer processes or threads to keep up with that
> load.
> > That seems quite excessive to me from a context switching and process
> > overhead perspective.
>
> Quick point:
> A single process using multiple threads dedicated to writing is an
> excellent optimization target on the win32 platform, (and if it is
> similarly useful on other platforms, so much the better).  To my way of
> thinking, this is an ideal approach in the long run.
>
> Multiple processes scheduling writes (even it is only 10), IMO, is a bad
> idea because of the way process management on win32 works for various
> reasons.

Yes, Win32 is going to need something like this because it doesn't have
sync.  The issue is whether Unix should use it too.

--
  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