You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@xalan.apache.org by Martin Sterman <ms...@exceloncorp.com> on 2000/08/18 16:57:05 UTC
Uni vs Quad Processor Performance
Just ran some tests using win32 Xalan C on
uniprocessor and quadprocessor NT machines.
This uses a compiled stylesheet to for-each iterate
through varying size documents.
Size uni quad
1 KB 15 ms 15 ms
10 30 30
100 312 1578
Subsequent iterations on the quad machine go even slower
(subsequent iterations improve performance on the uni).
Larger documents continue linear performance on the uni
and nonlinear performance on the quad.
Any ideas? Are there compiler options not set correctly?
RE: Uni vs Quad Processor Performance
Posted by Ed Staub <es...@mediaone.net>.
Martin,
Here's one wild hypothesis, a nasty one:
Is it possible that synchronized{} is very costly on your multiprocessor
box?
It looks like both Xerces and Xalan do quite a bit of it; I'm not
experienced enough in the code to be very reliable, though.
You might want to make a quick test of the cost of synchronized{} on one box
vs. the other, just to see if this is a possibility.
-Ed Staub
-----Original Message-----
From: Martin Sterman [mailto:msterman@exceloncorp.com]
Sent: Friday, August 18, 2000 12:38 PM
To: xalan-dev@xml.apache.org
Subject: Re: Uni vs Quad Processor Performance
Single thread. Document just gets wider
as it gets larger. Note the very linear performance
for the uniprocessor. Documents are the same for
both machines.
----- Original Message -----
From: "Ed Staub" <es...@mediaone.net>
To: <xa...@xml.apache.org>
Sent: Friday, August 18, 2000 11:44 AM
Subject: RE: Uni vs Quad Processor Performance
> Are you running multiple threads of this test, or a single thread?
>
> If it's just a single thread, this is _really_ weird.
> Is there something different in character about the 100KB content?
>
> -Ed Staub
>
> -----Original Message-----
> From: Martin Sterman [mailto:msterman@exceloncorp.com]
> Sent: Friday, August 18, 2000 10:57 AM
> To: xalan
> Subject: Uni vs Quad Processor Performance
>
>
> Just ran some tests using win32 Xalan C on
> uniprocessor and quadprocessor NT machines.
> This uses a compiled stylesheet to for-each iterate
> through varying size documents.
>
>
> Size uni quad
> 1 KB 15 ms 15 ms
> 10 30 30
> 100 312 1578
>
> Subsequent iterations on the quad machine go even slower
> (subsequent iterations improve performance on the uni).
> Larger documents continue linear performance on the uni
> and nonlinear performance on the quad.
>
> Any ideas? Are there compiler options not set correctly?
>
>
>
>
>
>
Re: Uni vs Quad Processor Performance
Posted by Martin Sterman <ms...@exceloncorp.com>.
Single thread. Document just gets wider
as it gets larger. Note the very linear performance
for the uniprocessor. Documents are the same for
both machines.
----- Original Message -----
From: "Ed Staub" <es...@mediaone.net>
To: <xa...@xml.apache.org>
Sent: Friday, August 18, 2000 11:44 AM
Subject: RE: Uni vs Quad Processor Performance
> Are you running multiple threads of this test, or a single thread?
>
> If it's just a single thread, this is _really_ weird.
> Is there something different in character about the 100KB content?
>
> -Ed Staub
>
> -----Original Message-----
> From: Martin Sterman [mailto:msterman@exceloncorp.com]
> Sent: Friday, August 18, 2000 10:57 AM
> To: xalan
> Subject: Uni vs Quad Processor Performance
>
>
> Just ran some tests using win32 Xalan C on
> uniprocessor and quadprocessor NT machines.
> This uses a compiled stylesheet to for-each iterate
> through varying size documents.
>
>
> Size uni quad
> 1 KB 15 ms 15 ms
> 10 30 30
> 100 312 1578
>
> Subsequent iterations on the quad machine go even slower
> (subsequent iterations improve performance on the uni).
> Larger documents continue linear performance on the uni
> and nonlinear performance on the quad.
>
> Any ideas? Are there compiler options not set correctly?
>
>
>
>
>
>
RE: Uni vs Quad Processor Performance
Posted by Ed Staub <es...@mediaone.net>.
Are you running multiple threads of this test, or a single thread?
If it's just a single thread, this is _really_ weird.
Is there something different in character about the 100KB content?
-Ed Staub
-----Original Message-----
From: Martin Sterman [mailto:msterman@exceloncorp.com]
Sent: Friday, August 18, 2000 10:57 AM
To: xalan
Subject: Uni vs Quad Processor Performance
Just ran some tests using win32 Xalan C on
uniprocessor and quadprocessor NT machines.
This uses a compiled stylesheet to for-each iterate
through varying size documents.
Size uni quad
1 KB 15 ms 15 ms
10 30 30
100 312 1578
Subsequent iterations on the quad machine go even slower
(subsequent iterations improve performance on the uni).
Larger documents continue linear performance on the uni
and nonlinear performance on the quad.
Any ideas? Are there compiler options not set correctly?