Re: Division of Load Average

Re: Division of Load Average

[sena seneviratne]

Arjan and Balbir,

Yes as discussed and agreed upon with Nagar and balbir  4 years ago this is 
how they are listed

Loadavg(1), Loadavg(5), Loadavg(15)
rootloadavg
uid_1    CPUload_1   DiskLoad_1
uid_2   CPUload_2   DiskLoad_2
uid_3   CPUload_3   DiskLoad_3

-----------------------------------------------

uid_n    CPUload_n   DiskLoad_n

Thanks
Sena Seneviratne



At 08:43 PM 10/1/2008 -0700, you wrote:
>On Thu, 2 Oct 2008 06:58:45 -0400
>auntvini@cel.usyd.edu.au wrote:
>
> > My self and my superviosr Dr David Levy have been working on
> > introduction of new kernel code to correct the inherent problem of
> > calculating load avarage by the current kernel. We have posted many
> > e-mails in this regard since 2003 and kernel has been updated under
> > the supervision of Rusty Russel.
> >
> > Also we had interactions with IBM group(balbir@in.ibm.com)  who was
> > engaged in delayed stat calculations.
> >
> > We emphasise the following facts
> > (1) seperation of Disk IO load from CPU load
> >
> > This comes as a correction to the existing kernel as the calculation
> > of the existing kernel load average is based on the arithmatic
> > addition of Disk IO and CPU loads
>
>while you can argue if including disk makes sense or not, it's 20+
>years of tradition on Linux, and more on Unix.
>
>So.. whatever you do (and I think there's some sense in splitting
>things out).. please only *add* stats, don't replace the existing ones.
>
>--
>Arjan van de Ven        Intel Open Source Technology Centre
>For development, discussion and tips for power savings,
>visit http://www.lesswatts.org

Re: Division of Load Average

[Arjan van de Ven]

On Thu, 2 Oct 2008 06:58:45 -0400
auntvini@cel.usyd.edu.au wrote:

> My self and my superviosr Dr David Levy have been working on
> introduction of new kernel code to correct the inherent problem of
> calculating load avarage by the current kernel. We have posted many
> e-mails in this regard since 2003 and kernel has been updated under
> the supervision of Rusty Russel.
> 
> Also we had interactions with IBM group(balbir@in.ibm.com)  who was
> engaged in delayed stat calculations.
> 
> We emphasise the following facts
> (1) seperation of Disk IO load from CPU load
> 
> This comes as a correction to the existing kernel as the calculation
> of the existing kernel load average is based on the arithmatic
> addition of Disk IO and CPU loads

while you can argue if including disk makes sense or not, it's 20+
years of tradition on Linux, and more on Unix.

So.. whatever you do (and I think there's some sense in splitting
things out).. please only *add* stats, don't replace the existing ones.

-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: Division of Load Average

[Balbir Singh]

sena seneviratne wrote:
> Arjan and Balbir,
> 
> Yes as discussed and agreed upon with Nagar and balbir  4 years ago this
> is how they are listed
> 
> Loadavg(1), Loadavg(5), Loadavg(15)
> rootloadavg
> uid_1    CPUload_1   DiskLoad_1
> uid_2   CPUload_2   DiskLoad_2
> uid_3   CPUload_3   DiskLoad_3
> 
> -----------------------------------------------
> 
> uid_n    CPUload_n   DiskLoad_n
> 

Sorry, I think I am completely out of context w.r.t. discussion. Any hints/URLs
to the old discussion would be nice.

-- 
	Balbir

Division of Load Average

[auntvini]

My self and my superviosr Dr David Levy have been working on introduction of new kernel code to correct the inherent problem of calculating load avarage by the current kernel. We have posted many e-mails in this regard since 2003 and kernel has been updated under the supervision of Rusty Russel.

Also we had interactions with IBM group(balbir@in.ibm.com)  who was engaged in delayed stat calculations.

We emphasise the following facts
(1) seperation of Disk IO load from CPU load

This comes as a correction to the existing kernel as the calculation of the existing kernel load average is based on the arithmatic addition of Disk IO and CPU loads

(2) In addition to that calculation of load averages for each current user
This is an extention we would like to suggest. For Grid and Cluster computing service providers can use the collection of user load profiles (CPU and Disk IO) for future predictions.

Thanks
Sena Seneviratne
Computer Engineering Lab
Sydney University
Australia

--
This message was sent on behalf of auntvini@cel.usyd.edu.au at openSubscriber.com
http://www.opensubscriber.com/messages/linux-kernel@vger.kernel.org/topic.html