2006-03-31 23:17:19

by Con Kolivas

[permalink] [raw]
Subject: Re: [ck] Re: Staircase test patch

On Saturday 01 April 2006 07:31, Thorsten Will wrote:
> On Friday 31 March 2006 23:07 +1000, Con Kolivas wrote:
> >Hi Thorsten et al
>
> Hi, Con.
>
> >Thorsten could you please test to see if this fixes the problem for you?
>
> Oh boy, oh boy, oh boy.
>
> Against a bash loop:
> |# dd bs=1M count=2048 </dev/hdb >/dev/null
> |2048+0 records in
> |2048+0 records out
> |2147483648 bytes transferred in 35.497603 seconds (60496582 bytes/sec)
>
> Yes! Success! And the crowd goes wild! :-)
>
> I think you finally nailed it. Thank you so much!

No, thank _you_ for bringing it to my attention and testing :)

Cheers,
Con


2006-04-01 00:22:54

by Peter Williams

[permalink] [raw]
Subject: Re: [ck] Re: Staircase test patch

Con Kolivas wrote:
> On Saturday 01 April 2006 07:31, Thorsten Will wrote:
>> On Friday 31 March 2006 23:07 +1000, Con Kolivas wrote:
>>> Hi Thorsten et al
>> Hi, Con.
>>
>>> Thorsten could you please test to see if this fixes the problem for you?
>> Oh boy, oh boy, oh boy.
>>
>> Against a bash loop:
>> |# dd bs=1M count=2048 </dev/hdb >/dev/null
>> |2048+0 records in
>> |2048+0 records out
>> |2147483648 bytes transferred in 35.497603 seconds (60496582 bytes/sec)
>>
>> Yes! Success! And the crowd goes wild! :-)
>>
>> I think you finally nailed it. Thank you so much!
>
> No, thank _you_ for bringing it to my attention and testing :)

Should I apply this to staircase in PlugSched?

Peter
--
Peter Williams [email protected]

"Learning, n. The kind of ignorance distinguishing the studious."
-- Ambrose Bierce

2006-04-01 00:32:04

by Con Kolivas

[permalink] [raw]
Subject: Re: Staircase test patch

On Saturday 01 April 2006 10:22, Peter Williams wrote:
> Con Kolivas wrote:
> > On Saturday 01 April 2006 07:31, Thorsten Will wrote:
> >> On Friday 31 March 2006 23:07 +1000, Con Kolivas wrote:
> >>> Hi Thorsten et al
> >>
> >> Hi, Con.
> >>
> >>> Thorsten could you please test to see if this fixes the problem for
> >>> you?
> >>
> >> Oh boy, oh boy, oh boy.
> >>
> >> Against a bash loop:
> >> |# dd bs=1M count=2048 </dev/hdb >/dev/null
> >> |2048+0 records in
> >> |2048+0 records out
> >> |2147483648 bytes transferred in 35.497603 seconds (60496582 bytes/sec)
> >>
> >> Yes! Success! And the crowd goes wild! :-)
> >>
> >> I think you finally nailed it. Thank you so much!
> >
> > No, thank _you_ for bringing it to my attention and testing :)
>
> Should I apply this to staircase in PlugSched?

I plan to make staircase v15 which is just this change, which would then need
to be resunc with plugsched. Unfortunately this needs code in
account_system_time which is not open to the schedulers in plugsched
currently so it needs more plugsched code to go in. I suspect other
schedulers may want to hook into this function, but I know you're currently
busy with smp nice to hack this in. I may look at doing it myself when I have
time if you don't have time.

Cheers,
Con

2006-04-01 00:45:11

by Peter Williams

[permalink] [raw]
Subject: Re: Staircase test patch

Con Kolivas wrote:
> On Saturday 01 April 2006 10:22, Peter Williams wrote:
>> Con Kolivas wrote:
>>> On Saturday 01 April 2006 07:31, Thorsten Will wrote:
>>>> On Friday 31 March 2006 23:07 +1000, Con Kolivas wrote:
>>>>> Hi Thorsten et al
>>>> Hi, Con.
>>>>
>>>>> Thorsten could you please test to see if this fixes the problem for
>>>>> you?
>>>> Oh boy, oh boy, oh boy.
>>>>
>>>> Against a bash loop:
>>>> |# dd bs=1M count=2048 </dev/hdb >/dev/null
>>>> |2048+0 records in
>>>> |2048+0 records out
>>>> |2147483648 bytes transferred in 35.497603 seconds (60496582 bytes/sec)
>>>>
>>>> Yes! Success! And the crowd goes wild! :-)
>>>>
>>>> I think you finally nailed it. Thank you so much!
>>> No, thank _you_ for bringing it to my attention and testing :)
>> Should I apply this to staircase in PlugSched?
>
> I plan to make staircase v15 which is just this change, which would then need
> to be resunc with plugsched. Unfortunately this needs code in
> account_system_time which is not open to the schedulers in plugsched
> currently so it needs more plugsched code to go in. I suspect other
> schedulers may want to hook into this function, but I know you're currently
> busy with smp nice to hack this in. I may look at doing it myself when I have
> time if you don't have time.

OK. I'm currently porting PlugSched to 2.6.16-mm2 which requires adding
priority inheritance to each scheduler. I'm modifying staircase and
nicksched myself but would appreciate a code review after I release it.

Peter
--
Peter Williams [email protected]

"Learning, n. The kind of ignorance distinguishing the studious."
-- Ambrose Bierce