ia_64_bit Performance difference

ia_64_bit Performance difference

[Nauman Tahir]

Hi all,
I have written a block device driver. Driver includes the
implementation of write back cache policy. Purpose of the driver is
not an issue. The problem I am facing is the considerable difference
of performance when I run same driver on 32 and 64 BIT OS. I am
testing the driver on 64 Bit Machine and run the same driver on both
(32 and 64 Bit) OS. On 32 Bit, IO rate is almost double then on 64 Bit
OS. ( i wish it could have been opposite :( )
Is there anything to do with 64-Bit kernel compilation issue. What
areas I have to look for?? I have used a few threads. Is there
anything to change related to thread priority or attributes??
regards
Nauman

Re: ia_64_bit Performance difference

[Arjan van de Ven]

On Tue, 2005-12-27 at 12:18 +0500, Nauman Tahir wrote:
> Hi all,
> I have written a block device driver. Driver includes the
> implementation of write back cache policy. Purpose of the driver is
> not an issue. The problem I am facing is the considerable difference
> of performance when I run same driver on 32 and 64 BIT OS. I am
> testing the driver on 64 Bit Machine and run the same driver on both
> (32 and 64 Bit) OS. On 32 Bit, IO rate is almost double then on 64 Bit
> OS. ( i wish it could have been opposite :( )

you forgot to post the URL to your source code...

Re: ia_64_bit Performance difference

[Nauman Tahir]

On 12/27/05, Arjan van de Ven <arjan@infradead.org> wrote:
> On Tue, 2005-12-27 at 12:18 +0500, Nauman Tahir wrote:
> > Hi all,
> > I have written a block device driver. Driver includes the
> > implementation of write back cache policy. Purpose of the driver is
> > not an issue. The problem I am facing is the considerable difference
> > of performance when I run same driver on 32 and 64 BIT OS. I am
> > testing the driver on 64 Bit Machine and run the same driver on both
> > (32 and 64 Bit) OS. On 32 Bit, IO rate is almost double then on 64 Bit
> > OS. ( i wish it could have been opposite :( )
>
> you forgot to post the URL to your source code...

I did NOT FORGET to post the URL of my code as I mentioned I am
interested in general
areas of the kernel to look for. Like is there any problem related to
64 bit specific kernel compilation etc...

You should ask for my system configurations. As far as the code is
concered there is not any architecture specific issue I guess in that.
I have used threads of MD as my generic thread library and have used
calculations based on unsigned long.

Kernel Version:
I am using is 2.6.9 and onwards.

Machine configuration is :
HP Prolient DL 145 (Dual Opteron 244 Processors with 14GB RAM)

Regards

Nauman

Re: ia_64_bit Performance difference

[Arjan van de Ven]

On Tue, 2005-12-27 at 13:10 +0500, Nauman Tahir wrote:
> On 12/27/05, Arjan van de Ven <arjan@infradead.org> wrote:
> > On Tue, 2005-12-27 at 12:18 +0500, Nauman Tahir wrote:
> > > Hi all,
> > > I have written a block device driver. Driver includes the
> > > implementation of write back cache policy. Purpose of the driver is
> > > not an issue. The problem I am facing is the considerable difference
> > > of performance when I run same driver on 32 and 64 BIT OS. I am
> > > testing the driver on 64 Bit Machine and run the same driver on both
> > > (32 and 64 Bit) OS. On 32 Bit, IO rate is almost double then on 64 Bit
> > > OS. ( i wish it could have been opposite :( )
> >
> > you forgot to post the URL to your source code...
> 
> I did NOT FORGET to post the URL of my code

it's gpl.. so what's the problem?
with the code we can see if you made a bug somewhere or did something
weird that will kill performance. without it... it's a guessing game
that only wastes our time. 

Re: ia_64_bit Performance difference

[Roland Dreier]

I would suggest you find the specific difference in the compiled
kernel and/or driver that is causing your driver to run slower and
then correct the issue.  That should fix your performance problems.

 - R.

Re: ia_64_bit Performance difference

[Bernd Eckenfels]

In article <f0309ff0512270010g39d42f83rf9da794f455854a5@mail.gmail.com> you wrote:
> HP Prolient DL 145 (Dual Opteron 244 Processors with 14GB RAM)

This is BTW not IA64 (as mentioned in subject)

Have you tried to profile it?

Gruss
Bernd

Re: ia_64_bit Performance difference

[Gerhard Mack]

On Tue, 27 Dec 2005, Nauman Tahir wrote:
> On 12/27/05, Arjan van de Ven <arjan@infradead.org> wrote:
> > On Tue, 2005-12-27 at 12:18 +0500, Nauman Tahir wrote:
> > > Hi all,
> > > I have written a block device driver. Driver includes the
> > > implementation of write back cache policy. Purpose of the driver is
> > > not an issue. The problem I am facing is the considerable difference
> > > of performance when I run same driver on 32 and 64 BIT OS. I am
> > > testing the driver on 64 Bit Machine and run the same driver on both
> > > (32 and 64 Bit) OS. On 32 Bit, IO rate is almost double then on 64 Bit
> > > OS. ( i wish it could have been opposite :( )
> >
> > you forgot to post the URL to your source code...
> 
> I did NOT FORGET to post the URL of my code as I mentioned I am
> interested in general
> areas of the kernel to look for. Like is there any problem related to
> 64 bit specific kernel compilation etc...
> 
> You should ask for my system configurations. As far as the code is
> concered there is not any architecture specific issue I guess in that.
> I have used threads of MD as my generic thread library and have used
> calculations based on unsigned long.
> 
> Kernel Version:
> I am using is 2.6.9 and onwards.
> 
> Machine configuration is :
> HP Prolient DL 145 (Dual Opteron 244 Processors with 14GB RAM)

Opteron is not ia_64 it's X86_64.  ia_64 is Itanuim.

	Gerhard




--
Gerhard Mack

gmack@innerfire.net

<>< As a computer I find your faith in technology amusing.