slow performance w/patch-2.6.7-mjb1

slow performance w/patch-2.6.7-mjb1

[Phy Prabab]

Hello,

To the mbligh, maintainer of mjb patch sets:

I am trying to track down why I am seeing 2x in run
time with patch-2.6.7-mjb1.  I would like to get the
4g/4g patch, hence the use of this patch set, however,
something within this patch has more than doubled the
run time for a test executable I have so I would like
to see what component might be the cause.  Is there a
list of the various patches that went into this patch
set and if so, are the patches in a broken out format?

Thanks!
Phy

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: slow performance w/patch-2.6.7-mjb1

[Martin J. Bligh]

> To the mbligh, maintainer of mjb patch sets:
> 
> I am trying to track down why I am seeing 2x in run
> time with patch-2.6.7-mjb1.  I would like to get the
> 4g/4g patch, hence the use of this patch set, however,
> something within this patch has more than doubled the
> run time for a test executable I have so I would like
> to see what component might be the cause.  Is there a
> list of the various patches that went into this patch
> set and if so, are the patches in a broken out format?

Yeah, though I'd start with "time foo" and if it's system time,
do a kernel profile. Let me know what it is, or give me a testcase ...
This is compared to 2.6.7 virgin? or -current? current code has some
wierd timer problems in mainline ... make sure to compare it to virgin.

Broken out patches are here under the patches/ subdir if you're desperate ;-)
eg: ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/patches/2.6.7/2.6.7-mjb1

Re: slow performance w/patch-2.6.7-mjb1

[Phy Prabab]

Martin et al,

So I configed with your patch just the basics and get
similar times that I do with 2.6.7 virigin and 2.4.21.
 However, as soon as I enable 4G split, the rt
increases by ~35s (out of 1m45s compared to 1m10s). 
Do you know if this is in line w/expectations?  Is
there anyway to reduce this?

Thank you for your time.
Phy

--- "Martin J. Bligh" <mbligh@aracnet.com> wrote:
> > To the mbligh, maintainer of mjb patch sets:
> > 
> > I am trying to track down why I am seeing 2x in
> run
> > time with patch-2.6.7-mjb1.  I would like to get
> the
> > 4g/4g patch, hence the use of this patch set,
> however,
> > something within this patch has more than doubled
> the
> > run time for a test executable I have so I would
> like
> > to see what component might be the cause.  Is
> there a
> > list of the various patches that went into this
> patch
> > set and if so, are the patches in a broken out
> format?
> 
> Yeah, though I'd start with "time foo" and if it's
> system time,
> do a kernel profile. Let me know what it is, or give
> me a testcase ...
> This is compared to 2.6.7 virgin? or -current?
> current code has some
> wierd timer problems in mainline ... make sure to
> compare it to virgin.
> 
> Broken out patches are here under the patches/
> subdir if you're desperate ;-)
> eg:
>
ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/patches/2.6.7/2.6.7-mjb1
> 
> -
> To unsubscribe from this list: send the line
> "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at 
> http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 


		
__________________________________
Do you Yahoo!?
Yahoo! Mail - 50x more storage than other providers!
http://promotions.yahoo.com/new_mail

Re: slow performance w/patch-2.6.7-mjb1

[Martin J. Bligh]

> So I configed with your patch just the basics and get
> similar times that I do with 2.6.7 virigin and 2.4.21.
>  However, as soon as I enable 4G split, the rt
> increases by ~35s (out of 1m45s compared to 1m10s). 
> Do you know if this is in line w/expectations?  Is
> there anyway to reduce this?

Syscalls, etc will definitely be slower ... but it's not normally
that severe ... what's the workload? And how much of hte increase
is systime vs user time? (use /usr/bin/time, not the shell builtin)

M.

Re: slow performance w/patch-2.6.7-mjb1

[Phy Prabab]

Martin et al:

Here is a litle bit more information:
2.6.7-mjb1 w/4G split enabled:
44.91user 56.95system 1:46.30elapsed 95%CPU
(0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+6907875minor)pagefaults
0swaps


2.6.7-mjb1 w/4G disabled enabled:
30.71user 34.56system 1:11.29elapsed 91%CPU
(0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (21major+6907525minor)pagefaults
0swaps

Clearly something is wrong.  This is making headers
which does a lot of spawning of bash shells and ln -s
different files and some minor dependancy makes.

Any help understanding what is happending here would
be greatly appreciated!

Thanks!
Phy


--- "Martin J. Bligh" <mbligh@aracnet.com> wrote:
> > So I configed with your patch just the basics and
> get
> > similar times that I do with 2.6.7 virigin and
> 2.4.21.
> >  However, as soon as I enable 4G split, the rt
> > increases by ~35s (out of 1m45s compared to
> 1m10s). 
> > Do you know if this is in line w/expectations?  Is
> > there anyway to reduce this?
> 
> Syscalls, etc will definitely be slower ... but it's
> not normally
> that severe ... what's the workload? And how much of
> hte increase
> is systime vs user time? (use /usr/bin/time, not the
> shell builtin)
> 
> M.
> 
> 
> 
> 



	
		
__________________________________
Do you Yahoo!?
New and Improved Yahoo! Mail - 100MB free storage!
http://promotions.yahoo.com/new_mail 

Re: slow performance w/patch-2.6.7-mjb1

[Martin J. Bligh]

> Here is a litle bit more information:
> 2.6.7-mjb1 w/4G split enabled:
> 44.91user 56.95system 1:46.30elapsed 95%CPU
> (0avgtext+0avgdata 0maxresident)k
> 0inputs+0outputs (0major+6907875minor)pagefaults
> 0swaps
> 
> 
> 2.6.7-mjb1 w/4G disabled enabled:
> 30.71user 34.56system 1:11.29elapsed 91%CPU
> (0avgtext+0avgdata 0maxresident)k
> 0inputs+0outputs (21major+6907525minor)pagefaults
> 0swaps
> 
> Clearly something is wrong.  This is making headers
> which does a lot of spawning of bash shells and ln -s
> different files and some minor dependancy makes.
> 
> Any help understanding what is happending here would
> be greatly appreciated!

Mmmm. Try grabbing a kernel profile of both, and it might
be easier to see what the overhead is, exactly. See 
Documentation/basic_profiling.txt for instructions if you haven't
done it before.

M.