atropos still broken?

atropos still broken?

[Diwaker Gupta]

Hi everyone,

This is for Xen 2.0.3. I'm really interested in the functionality of
atropos (enforcing CPU partitioning in a non work-conserving fashion).
As I had reported earlier, atropos seemed broken in 2.0. It still
seems broken now:

$ xm list
Name              Id  Mem(MB)  CPU  State  Time(s)  Console
Domain-0           0      123    0  r----     49.1
vm1                1       47    0  -b---     29.7    9601

$ xm atropos 0 10000 100000 50000 1
$ xm atropos 0 70000 100000 50000 1

Now if I run the slurp program (posted earlier in this list) in vm1,
it eats up all the CPU. If I start the same program in dom0 as well,
both get ~50% of the CPU. With those parameters, ideally vm1 should be
getting 70% and no more, and dom0 should be getting 10% and no more.

So it seems atropos is still broken. Is a fix being worked upon? I'd
really like to help in any way I can. I'm familiar with the Xen
scheduling code; if any of the developers have any idea of where/what
the bug is, I'd like to take a crack at it.

Thanks,
-- 
Diwaker Gupta
http://resolute.ucsd.edu/diwaker


-------------------------------------------------------
This SF.Net email is sponsored by: IntelliVIEW -- Interactive Reporting
Tool for open source databases. Create drag-&-drop reports. Save time
by over 75%! Publish reports on the web. Export to DOC, XLS, RTF, etc.
Download a FREE copy at http://www.intelliview.com/go/osdn_nl

Re: atropos still broken?

[Steven Hand]

>So it seems atropos is still broken. Is a fix being worked upon? I'd
>really like to help in any way I can. I'm familiar with the Xen
>scheduling code; if any of the developers have any idea of where/what
>the bug is, I'd like to take a crack at it.

Stephan Dielstelhorst is working on a clean implementation which I 
believe is close to ready for testing now. We'll probably aim to 
push this to 2.0-testing in the near future, but if people are 
interested in 'beta test' we can probably email a patch a bit earlier. 

cheers,

S.


-------------------------------------------------------
This SF.Net email is sponsored by: IntelliVIEW -- Interactive Reporting
Tool for open source databases. Create drag-&-drop reports. Save time
by over 75%! Publish reports on the web. Export to DOC, XLS, RTF, etc.
Download a FREE copy at http://www.intelliview.com/go/osdn_nl

Re: atropos still broken?

[Gregor Milos]

> Hi everyone,
>
> This is for Xen 2.0.3. I'm really interested in the functionality of
> atropos (enforcing CPU partitioning in a non work-conserving fashion).
> As I had reported earlier, atropos seemed broken in 2.0. It still
> seems broken now:
>
> $ xm list
> Name              Id  Mem(MB)  CPU  State  Time(s)  Console
> Domain-0           0      123    0  r----     49.1
> vm1                1       47    0  -b---     29.7    9601
>
> $ xm atropos 0 10000 100000 50000 1
> $ xm atropos 0 70000 100000 50000 1
>
> Now if I run the slurp program (posted earlier in this list) in vm1,
> it eats up all the CPU. If I start the same program in dom0 as well,
> both get ~50% of the CPU. With those parameters, ideally vm1 should be
> getting 70% and no more, and dom0 should be getting 10% and no more.

I do not remember exactly what order are the 5 parameters in, so I am not 
quite certain what you are trying to set. Anyhow the Atropos is a kind of 
earliest deadline first scheduler that tries to fullfill the guarantees 
first, but if there is any spare CPU time left it will do best effort 
scheduling (that is the reason dom0 is getting 100% CPU when you run slurp in 
it).

> So it seems atropos is still broken. Is a fix being worked upon? I'd

I do not think anybody is working on it at the moment (Mark, as I recall you 
were the last one to play with Atropos, can you possibly give more details?), 
but since I am familiar with the scheduling code (although I have never 
worked on Atropos really), I am keen on tackling this problem.

> really like to help in any way I can. I'm familiar with the Xen
> scheduling code; if any of the developers have any idea of where/what
> the bug is, I'd like to take a crack at it.

That's great, we can try to fix it together than (I contact you offlist 
tomorrow).

Cheers
Gregor

-- 
Quidquid latine dictum sit, altum viditur --- Anon


-------------------------------------------------------
This SF.Net email is sponsored by: IntelliVIEW -- Interactive Reporting
Tool for open source databases. Create drag-&-drop reports. Save time
by over 75%! Publish reports on the web. Export to DOC, XLS, RTF, etc.
Download a FREE copy at http://www.intelliview.com/go/osdn_nl

Re: atropos still broken?

[Mark Williamson]

> This is for Xen 2.0.3. I'm really interested in the functionality of
> atropos (enforcing CPU partitioning in a non work-conserving fashion).
> As I had reported earlier, atropos seemed broken in 2.0. It still
> seems broken now:

Yes - it's still broken because I still haven't fixed it :-(

> So it seems atropos is still broken. Is a fix being worked upon? I'd

Fixing Atropos is still on my todo list.  One of the other developers is 
working on an alternative deadline-based scheduler, which may behave better.

> scheduling code; if any of the developers have any idea of where/what
> the bug is, I'd like to take a crack at it.

I don't know where the bug is - it appears to be something quite subtle since 
the code *looks* OK to the people who have read through it...  Analysing some 
trace buffer output might help us figure out what it's actually doing and 
why.

I'd still like to look at this myself but you're welcome to take a shot at 
debugging it if you're feeling enthusiastic!

Cheers,
Mark


-------------------------------------------------------
This SF.Net email is sponsored by: IntelliVIEW -- Interactive Reporting
Tool for open source databases. Create drag-&-drop reports. Save time
by over 75%! Publish reports on the web. Export to DOC, XLS, RTF, etc.
Download a FREE copy at http://www.intelliview.com/go/osdn_nl

Re: atropos still broken?

[Stephan Diestelhorst]

> > So it seems atropos is still broken. Is a fix being worked upon? I'd
>
> Fixing Atropos is still on my todo list.  One of the other developers is
> working on an alternative deadline-based scheduler, which may behave
better.
>

Indeed, that's me :) My scheduler is working quite well, but I want to do
some testing/benchmarking today before it gets released to the "wild"...



-------------------------------------------------------
This SF.Net email is sponsored by: IntelliVIEW -- Interactive Reporting
Tool for open source databases. Create drag-&-drop reports. Save time
by over 75%! Publish reports on the web. Export to DOC, XLS, RTF, etc.
Download a FREE copy at http://www.intelliview.com/go/osdn_nl