atropos again

atropos again

[xuehai zhang]

Hi all,
I read the previous threads in the mailing list (I attached Ross's post sent on 2/23/05) about 
atropos scheduler and it looks to me the bug in atropos has not been fixed until that time. Because 
I could not find any new posts talking about atropos, I appreciate if anyone can update me the 
current status of the atropos scheduler.
BTW, Ross also mentioned Stephan is working on an alternative scheduler and I wonder where I can get 
the patch or is it already included in the latest version of Xen?
Thank you so much for your help.
Xuehai


 > Hi,
 >
 > I am working on a project which requires enforced CPU partitioning
 > between domains, and so was planning to use the atropos scheduler.
 > However I notice from a previous post that atropos does not seem to be
 > working.  Has there been any progress in finding the bug?
 >
 > I also noticed that Stephan Diestelhorst was working on an alternative
 > deadline-based scheduler.  Is this ready for use?
 >
 >
 > Thanks for your help
 >
 > Ross
 >

Re: atropos again

[xuehai zhang]

Hi,
Is there anyway (some command I type in the commandline) I can find which scheudler is currently 
used by Xen?
BTW, Could anyone answer my question aobut the current status of Atropos scheduler I sent a couple 
of days ago?
Many thanks.
Xuehai

xuehai zhang wrote:
> Hi all,
> I read the previous threads in the mailing list (I attached Ross's post 
> sent on 2/23/05) about atropos scheduler and it looks to me the bug in 
> atropos has not been fixed until that time. Because I could not find any 
> new posts talking about atropos, I appreciate if anyone can update me 
> the current status of the atropos scheduler.
> BTW, Ross also mentioned Stephan is working on an alternative scheduler 
> and I wonder where I can get the patch or is it already included in the 
> latest version of Xen?
> Thank you so much for your help.
> Xuehai
> 
> 
>  > Hi,
>  >
>  > I am working on a project which requires enforced CPU partitioning
>  > between domains, and so was planning to use the atropos scheduler.
>  > However I notice from a previous post that atropos does not seem to be
>  > working.  Has there been any progress in finding the bug?
>  >
>  > I also noticed that Stephan Diestelhorst was working on an alternative
>  > deadline-based scheduler.  Is this ready for use?
>  >
>  >
>  > Thanks for your help
>  >
>  > Ross
>  >
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
> 

Re: atropos again

[Stephan Diestelhorst]

> Hi,
> Is there anyway (some command I type in the commandline) I can find
> which scheudler is currently used by Xen?
You might wish to have a look at the boot output by Xen, either at an
attached serial console or by 'xm dmsg'. It tells you which scheduler
was used during the startup of the machine.

> BTW, Could anyone answer my question aobut the current status of Atropos
> scheduler I sent a couple of days ago
I've ported the scheduler to the unstable tree, and it will hopefully
get checked in in the near future! (*prod*)

Cheers,
   Stephan

Re: atropos again

[Xuehai Zhang]

Stephan,

Thanks for the reply.

> > Is there anyway (some command I type in the commandline) I can find
> > which scheudler is currently used by Xen?
> You might wish to have a look at the boot output by Xen, either at an
> attached serial console or by 'xm dmsg'. It tells you which scheduler
> was used during the startup of the machine.

Understood.

> > BTW, Could anyone answer my question aobut the current status of Atropos
> > scheduler I sent a couple of days ago
> I've ported the scheduler to the unstable tree, and it will hopefully
> get checked in in the near future! (*prod*)

The scheduler you mean is the atropos replacement you develop or just
atropos? If it is the former, what is the name of the replacement? Can I
just use "sched=<name>" as what I use for other scheduler to enable it?
Sinced it has been ported to the unstable tree, can I assume the testing
tree has been ported as well?

Thanks again for the help.

Xuehai

Re: atropos again

[Stephan Diestelhorst]

>>>BTW, Could anyone answer my question aobut the current status of Atropos
>>>scheduler I sent a couple of days ago
>>
>>I've ported the scheduler to the unstable tree, and it will hopefully
>>get checked in in the near future! (*prod*)
> 
> 
> The scheduler you mean is the atropos replacement you develop or just
> atropos? If it is the former, what is the name of the replacement? Can I
> just use "sched=<name>" as what I use for other scheduler to enable it?
> Sinced it has been ported to the unstable tree, can I assume the testing
> tree has been ported as well?

It is quite versatile, you can compile it so that it just does Atropos,
but it can do much more ;-) The current name for the thing is 'sedf'
which originated from Simple-EDF, but probably it's time now to call it
'Smart-EDF' or come up with a more reasonable name in general!

As it is not yet in the repositories you can't use it though...

ATM there is also a version for the testing tree, but as I said they are
not in the main repos yet.

Cheers,
   Stephan

Re: atropos again

[xuehai zhang]

Stephan,

Thanks for the response.

>>>>BTW, Could anyone answer my question aobut the current status of Atropos
>>>>scheduler I sent a couple of days ago
>>>
>>>I've ported the scheduler to the unstable tree, and it will hopefully
>>>get checked in in the near future! (*prod*)
>>
>>
>>The scheduler you mean is the atropos replacement you develop or just
>>atropos? If it is the former, what is the name of the replacement? Can I
>>just use "sched=<name>" as what I use for other scheduler to enable it?
>>Sinced it has been ported to the unstable tree, can I assume the testing
>>tree has been ported as well?
> 
> 
> It is quite versatile, you can compile it so that it just does Atropos,
> but it can do much more ;-) The current name for the thing is 'sedf'
> which originated from Simple-EDF, but probably it's time now to call it
> 'Smart-EDF' or come up with a more reasonable name in general!
> 
> As it is not yet in the repositories you can't use it though...


> ATM there is also a version for the testing tree, but as I said they are
> not in the main repos yet.

To clarify: even if I use "bk clone bk://xen.bkbits.net/xen-2.0-testing.bk" to get the latest 
testing tree, I will not find sedf scheduler because it has not been included yet.

It is still unclear to me if Atropos scheduler in the above testing tree works correctly or still 
buggy as pointed out by previous posts 2 month ago.

Thanks again for your help.

Xuehai

> Cheers,
>    Stephan
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
> 

Re: atropos again

[Stephan Diestelhorst]

>> As it is not yet in the repositories you can't use it though...
> 
> 
> 
>> ATM there is also a version for the testing tree, but as I said they are
>> not in the main repos yet.
> 
> 
> To clarify: even if I use "bk clone
> bk://xen.bkbits.net/xen-2.0-testing.bk" to get the latest testing tree,
> I will not find sedf scheduler because it has not been included yet.

Exactly.

> It is still unclear to me if Atropos scheduler in the above testing tree
> works correctly or still buggy as pointed out by previous posts 2 month
> ago.

As far as I know it is still broken, though you might want to try for
yourself.

Best,
   Stephan

Re: atropos again

[Stephan Diestelhorst]

> Stephan,
> Thanks a lot for sharing the SEDF patch. BTW, where can I find the
> reference to use it?

SEDF patch now in the public unstable tree, too! Check it out!
There should be a mini-HOWTO, with some examples.

> Is the usage similar to Atropos (for example, "xm
> atropos DOM PERIOD SLICE LATENCY XTRATIME")?

xm sedf DOM PERIOD SLICE LATENCY XTRATIME WEIGHT

Where:
PERIOD, SLICE: times for guarantees (NANOseconds)
LATENCY: currently ignored, recompile with other #define's in
sched_sedf.c (ns)
XTRATIME: flag, set to 1 and domain gets idle cpu time additional to
it's share
WEIGHT: use, if you don't want to specify periods/slices, simply
distributes the cpu according to these weights

Example:

xm sedf 1 20000000  5000000 0 0 0
xm sedf 2 25000000 12500000 0 0 0

-dom 1 gets 25% cpu share (5 ms every 20 ms)
-dom 2 gets 50% cpu share (12,5 ms every 25 ms)

xm sedf 1 0 0 0 0 0 1
xm sedf 2 0 0 0 0 0 5

-dom 2 gets 5-times more cpu than dom 1
-in total they get 95% - utilisation of fully specified domains

Hope this helps,
  Stephan

> On Mon, 9 May 2005, Stephan Diestelhorst wrote:
> 
> 
>>Try this one, it's using the 2.05 repo...
>>
>>Cheers,
>>  Stephan
>>
>>>Stephan,
>>>
>>>I wonder if it is possible that you can provide me a beta version of SEDF
>>>because it is not in the repository yet. I just start a project to explore
>>>resource enforcement in Xen and it will be very helpful that I can have the
>>>working SEDF.
>>>
>>>Thanks in advance for your help.
>>>
>>>Xuehai
>>>
>>>
>>>>>>As it is not yet in the repositories you can't use it though...
>>>>>>
>>>>>>
>>>>>>
>>>>>>ATM there is also a version for the testing tree, but as I said they are
>>>>>>not in the main repos yet.
>>>>>
>>>>>To clarify: even if I use "bk clone
>>>>>bk://xen.bkbits.net/xen-2.0-testing.bk" to get the latest testing tree,
>>>>>I will not find sedf scheduler because it has not been included yet.
>>>>
>>>>Exactly.
>>>>
>>>>
>>>>>It is still unclear to me if Atropos scheduler in the above testing tree
>>>>>works correctly or still buggy as pointed out by previous posts 2 month
>>>>>ago.
>>>>
>>>>As far as I know it is still broken, though you might want to try for
>>>>yourself.
>>>>
>>>>Best,
>>>>   Stephan
>>

Re: atropos again

[Xuehai Zhang]

> > Stephan,
> > Thanks a lot for sharing the SEDF patch. BTW, where can I find the
> > reference to use it?
>
> SEDF patch now in the public unstable tree, too! Check it out!
> There should be a mini-HOWTO, with some examples.

Thanks. I will check the unstable tree out!

>
> > Is the usage similar to Atropos (for example, "xm
> > atropos DOM PERIOD SLICE LATENCY XTRATIME")?
>
> xm sedf DOM PERIOD SLICE LATENCY XTRATIME WEIGHT
>
> Where:
> PERIOD, SLICE: times for guarantees (NANOseconds)
> LATENCY: currently ignored, recompile with other #define's in
> sched_sedf.c (ns)
> XTRATIME: flag, set to 1 and domain gets idle cpu time additional to
> it's share
> WEIGHT: use, if you don't want to specify periods/slices, simply
> distributes the cpu according to these weights
>
> Example:
>
> xm sedf 1 20000000  5000000 0 0 0
> xm sedf 2 25000000 12500000 0 0 0
>
> -dom 1 gets 25% cpu share (5 ms every 20 ms)
> -dom 2 gets 50% cpu share (12,5 ms every 25 ms)
>
> xm sedf 1 0 0 0 0 0 1
> xm sedf 2 0 0 0 0 0 5
>
> -dom 2 gets 5-times more cpu than dom 1
> -in total they get 95% - utilisation of fully specified domains

I can understand dom2 get 5 times more cpu than dom1, but why in total
they get 95%, not 94% or 96%?

Thanks.

Xuehai

Re: atropos again

[Stephan Diestelhorst]

> > xm sedf DOM PERIOD SLICE LATENCY XTRATIME WEIGHT
> >
> > Where:
> > PERIOD, SLICE: times for guarantees (NANOseconds)
> > LATENCY: currently ignored, recompile with other #define's in
> > sched_sedf.c (ns)
> > XTRATIME: flag, set to 1 and domain gets idle cpu time additional to
> > it's share
> > WEIGHT: use, if you don't want to specify periods/slices, simply
> > distributes the cpu according to these weights
> >
> > Example:
> >
> > xm sedf 1 20000000  5000000 0 0 0
> > xm sedf 2 25000000 12500000 0 0 0
> >
> > -dom 1 gets 25% cpu share (5 ms every 20 ms)
> > -dom 2 gets 50% cpu share (12,5 ms every 25 ms)
> >
> > xm sedf 1 0 0 0 0 0 1
> > xm sedf 2 0 0 0 0 0 5
> >
> > -dom 2 gets 5-times more cpu than dom 1
> > -in total they get 95% - utilisation of fully specified domains
>
> I can understand dom2 get 5 times more cpu than dom1, but why in total
> they get 95%, not 94% or 96%?

This is because I leave 5% unused, just in case :) This is of course 
changeable in sched_sedf.c.

Cheers,
   Stephan