Plans for better performance metrics in upcoming kernels?

Plans for better performance metrics in upcoming kernels?

[Scott Chapman]

Hi,
I'm wondering what the plans are for more accurate and more useful performance 
metrics in upcoming kernels.

CPU Utilization by process is apparently a known-inaccuracy.

There are no disk I/O metrics per process.

CPU Queue Length doesn't appear to be available?

Etc.

Linux clearly falls behind the competition in this area. It makes it rather 
tough to do system performance analysis on a Linux box!  :-)

Is there a plan to deal with these issues?  ETA's?

Cordially,
Scott

Re: Plans for better performance metrics in upcoming kernels?

[Nick Piggin]

Scott Chapman wrote:

>Hi,
>I'm wondering what the plans are for more accurate and more useful performance 
>metrics in upcoming kernels.
>
>CPU Utilization by process is apparently a known-inaccuracy.
>

This could be improved using Ingo's nanosecond scheduler patch.

>
>There are no disk I/O metrics per process.
>

This is now quite easy with per process IO contexts.

>
>CPU Queue Length doesn't appear to be available?
>

It wouldn't be difficult.

>
>Etc.
>
>Linux clearly falls behind the competition in this area. It makes it rather 
>tough to do system performance analysis on a Linux box!  :-)
>
>Is there a plan to deal with these issues?  ETA's?
>
>

Thinks are worked on depending on demand, and interest. I think a lot
of people are put of doing good kernel metrics due to lack of good
extensible userspace tools, and maybe a lack of standard ways to do the
exporting.

Re: Plans for better performance metrics in upcoming kernels?

[Scott Chapman]

Hi all,

I received one reply to this email.  I take it there is nobody really heading 
up the implementation of improved/missing performance metrics in the kernel?

Is this on anyone's radar?

What is Linus' take on more/better metrics?  
Is he likely to roll them in if they are implemented?

Is there any plan to implement any existing patches? 

If I can get developer time applied to this project, I don't want to re-invent 
the existing pieces.  Who do I coordinate with?

Cordially,
Scott

On Friday 05 September 2003 16:41, Scott Chapman wrote:
> Hi,
> I'm wondering what the plans are for more accurate and more useful
> performance metrics in upcoming kernels.
>
> CPU Utilization by process is apparently a known-inaccuracy.
>
> There are no disk I/O metrics per process.
>
> CPU Queue Length doesn't appear to be available?
>
> Etc.
>
> Linux clearly falls behind the competition in this area. It makes it rather
> tough to do system performance analysis on a Linux box!  :-)
>
> Is there a plan to deal with these issues?  ETA's?
>
> Cordially,
> Scott

Re: Plans for better performance metrics in upcoming kernels?

[Craig Thomas]

We have a project to maintain and improve systat here at OSDL.  We're 
looking for others to help in this area, if your're interested. 

On Mon, 2003-09-08 at 07:54, Scott Chapman wrote:
> Hi all,
> 
> I received one reply to this email.  I take it there is nobody really heading 
> up the implementation of improved/missing performance metrics in the kernel?
> 
> Is this on anyone's radar?
> 
> What is Linus' take on more/better metrics?  
> Is he likely to roll them in if they are implemented?
> 
> Is there any plan to implement any existing patches? 
> 
> If I can get developer time applied to this project, I don't want to re-invent 
> the existing pieces.  Who do I coordinate with?
> 
> Cordially,
> Scott
> 
> On Friday 05 September 2003 16:41, Scott Chapman wrote:
> > Hi,
> > I'm wondering what the plans are for more accurate and more useful
> > performance metrics in upcoming kernels.
> >
> > CPU Utilization by process is apparently a known-inaccuracy.
> >
> > There are no disk I/O metrics per process.
> >
> > CPU Queue Length doesn't appear to be available?
> >
> > Etc.
> >
> > Linux clearly falls behind the competition in this area. It makes it rather
> > tough to do system performance analysis on a Linux box!  :-)
> >
> > Is there a plan to deal with these issues?  ETA's?
> >
> > Cordially,
> > Scott
> 
> -
> 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/
-- 
Craig Thomas
craiger@osdl.org

Re: Plans for better performance metrics in upcoming kernels?

[Cliff White]

> We have a project to maintain and improve systat here at OSDL.  We're 
> looking for others to help in this area, if your're interested. 
> 
> On Mon, 2003-09-08 at 07:54, Scott Chapman wrote:
> > Hi all,
> > 
> > I received one reply to this email.  I take it there is nobody really heading 
> > up the implementation of improved/missing performance metrics in the kernel?
> > 
> > Is this on anyone's radar?
> > 
> > What is Linus' take on more/better metrics?  
> > Is he likely to roll them in if they are implemented?
> > 
> > Is there any plan to implement any existing patches? 
> > 
> > If I can get developer time applied to this project, I don't want to re-invent 
> > the existing pieces.  Who do I coordinate with?

OSDL can also provide you hardware resources, if you want to do a project.
We may be able to help some with the coordination, if bandwidth allows.
OSDL is definately interested in improving performance metrics.
cliffw

> > 
> > Cordially,
> > Scott
> > 
> > On Friday 05 September 2003 16:41, Scott Chapman wrote:
> > > Hi,
> > > I'm wondering what the plans are for more accurate and more useful
> > > performance metrics in upcoming kernels.
> > >
> > > CPU Utilization by process is apparently a known-inaccuracy.
> > >
> > > There are no disk I/O metrics per process.
> > >
> > > CPU Queue Length doesn't appear to be available?
> > >
> > > Etc.
> > >
> > > Linux clearly falls behind the competition in this area. It makes it rather
> > > tough to do system performance analysis on a Linux box!  :-)
> > >
> > > Is there a plan to deal with these issues?  ETA's?
> > >
> > > Cordially,
> > > Scott
> > 
> > -
> > 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/
> -- 
> Craig Thomas
> craiger@osdl.org
> 
> -
> 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/
> 

Re: Plans for better performance metrics in upcoming kernels?

[Peter Chubb]

>>>>> "Scott" == Scott Chapman <scott_list@mischko.com> writes:

Scott> Hi all, I received one reply to this email.  I take it there is
Scott> nobody really heading up the implementation of improved/missing
Scott> performance metrics in the kernel?


I've been working on microstate acounting, and am interested in better
metrics overall, for capacity planning and for accounting.

I think that SGI also have been doing some work -- see
http://oss.sgi.com/projects/csa/ 

Peter C