The "best" value of HZ

The "best" value of HZ

[Claudio Scordino]

Hi,

    during the last years there has been a lot of discussion about the "best" 
value of HZ... On i386 was 100, then became 1000, and finally was set to 250.
I'm thinking to do an evaluation of this parameter using different 
architectures.

Has anybody thought to give the possibility to modify the value of HZ at boot 
time instead of at compile time ? This would allow to easily test different 
values on different machines and create a table containing the "best" value 
for each architecture...  At this moment, instead, we have to recompile the 
kernel for each different value :(

Do you think there would be much work to do that ? 
Do you think it would be a desired feature the knowledge of the best value for 
each architecture with more precision ?

Thanks,

      Claudio

Re: The "best" value of HZ

[Luke Yang]

  yes,  our Blackfin CPU (from Analog Device) is team interested in
it.  Dynamically modify HZ will be useful when we evaluate the
performance.

On 10/28/05, Claudio Scordino <cloud.of.andor@gmail.com> wrote:
> Hi,
>
>     during the last years there has been a lot of discussion about the "best"
> value of HZ... On i386 was 100, then became 1000, and finally was set to 250.
> I'm thinking to do an evaluation of this parameter using different
> architectures.
>
> Has anybody thought to give the possibility to modify the value of HZ at boot
> time instead of at compile time ? This would allow to easily test different
> values on different machines and create a table containing the "best" value
> for each architecture...  At this moment, instead, we have to recompile the
> kernel for each different value :(
>
> Do you think there would be much work to do that ?
> Do you think it would be a desired feature the knowledge of the best value for
> each architecture with more precision ?
>
> Thanks,
>
>       Claudio
> -
> 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: The "best" value of HZ

[Randy.Dunlap]

On Fri, 28 Oct 2005 01:18:41 +0200 Claudio Scordino wrote:

> Hi,
> 
>     during the last years there has been a lot of discussion about the "best" 
> value of HZ... On i386 was 100, then became 1000, and finally was set to 250.
> I'm thinking to do an evaluation of this parameter using different 
> architectures.
> 
> Has anybody thought to give the possibility to modify the value of HZ at boot 
> time instead of at compile time ? This would allow to easily test different 
> values on different machines and create a table containing the "best" value 
> for each architecture...  At this moment, instead, we have to recompile the 
> kernel for each different value :(
> 
> Do you think there would be much work to do that ? 

Not a lot.  Could be useful.

> Do you think it would be a desired feature the knowledge of the best value for 
> each architecture with more precision ?

In this thread (around 2005-july-14)
Subject: Re: [PATCH] i386: Selectable Frequency of the Timer Interrupt
there was some interest in that, and the beginnings of a patch.
But the dynamic tick patch seems to have sidelined this one.

---
~Randy

Re: The "best" value of HZ

[Alistair John Strachan]

On Friday 28 October 2005 00:18, Claudio Scordino wrote:
> Hi,
>
>     during the last years there has been a lot of discussion about the
> "best" value of HZ... On i386 was 100, then became 1000, and finally was
> set to 250. I'm thinking to do an evaluation of this parameter using
> different architectures.
>
> Has anybody thought to give the possibility to modify the value of HZ at
> boot time instead of at compile time ? This would allow to easily test
> different values on different machines and create a table containing the
> "best" value for each architecture...  At this moment, instead, we have to
> recompile the kernel for each different value :(
>
> Do you think there would be much work to do that ?
> Do you think it would be a desired feature the knowledge of the best value
> for each architecture with more precision ?

Google for "dynticks". There's obviously an overhead associated with HZ not 
being a constant (the compiler cannot optimise many expressions), but the 
feature is being worked on nonetheless.

-- 
Cheers,
Alistair.

'No sense being pessimistic, it probably wouldn't work anyway.'
Third year Computer Science undergraduate.
1F2 55 South Clerk Street, Edinburgh, UK.

Re: The "best" value of HZ

[Lee Revell]

On Fri, 2005-10-28 at 03:31 +0100, Alistair John Strachan wrote:
> On Friday 28 October 2005 00:18, Claudio Scordino wrote:
> > Hi,
> >
> >     during the last years there has been a lot of discussion about the
> > "best" value of HZ... On i386 was 100, then became 1000, and finally was
> > set to 250. I'm thinking to do an evaluation of this parameter using
> > different architectures.
> >
> > Has anybody thought to give the possibility to modify the value of HZ at
> > boot time instead of at compile time ? This would allow to easily test
> > different values on different machines and create a table containing the
> > "best" value for each architecture...  At this moment, instead, we have to
> > recompile the kernel for each different value :(
> >
> > Do you think there would be much work to do that ?
> > Do you think it would be a desired feature the knowledge of the best value
> > for each architecture with more precision ?
> 
> Google for "dynticks". There's obviously an overhead associated with HZ not 
> being a constant (the compiler cannot optimise many expressions), but the 
> feature is being worked on nonetheless.
> 

Well Linus had the best idea in that thread (as usual) which was to
implement "dynamic ticks" by leaving HZ a constant, setting it to a high
value, and skipping ticks when idle.  Has there been any work in that
direction?

Lee

Re: The "best" value of HZ

[Jesper Juhl]

On 10/28/05, Lee Revell <rlrevell@joe-job.com> wrote:
> On Fri, 2005-10-28 at 03:31 +0100, Alistair John Strachan wrote:
> > On Friday 28 October 2005 00:18, Claudio Scordino wrote:
> > > Hi,
> > >
> > >     during the last years there has been a lot of discussion about the
> > > "best" value of HZ... On i386 was 100, then became 1000, and finally was
> > > set to 250. I'm thinking to do an evaluation of this parameter using
> > > different architectures.
> > >
> > > Has anybody thought to give the possibility to modify the value of HZ at
> > > boot time instead of at compile time ? This would allow to easily test
> > > different values on different machines and create a table containing the
> > > "best" value for each architecture...  At this moment, instead, we have to
> > > recompile the kernel for each different value :(
> > >
> > > Do you think there would be much work to do that ?
> > > Do you think it would be a desired feature the knowledge of the best value
> > > for each architecture with more precision ?
> >
> > Google for "dynticks". There's obviously an overhead associated with HZ not
> > being a constant (the compiler cannot optimise many expressions), but the
> > feature is being worked on nonetheless.
> >
>
> Well Linus had the best idea in that thread (as usual) which was to
> implement "dynamic ticks" by leaving HZ a constant, setting it to a high
> value, and skipping ticks when idle.  Has there been any work in that
> direction?
>

i did a bit of work in that area, but the stuff I came up with never
seemed to work right, so I dropped it.


--
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html

Re: The "best" value of HZ

[Con Kolivas]

On Fri, 28 Oct 2005 02:00 pm, Jesper Juhl wrote:
> On 10/28/05, Lee Revell <rlrevell@joe-job.com> wrote:
> > On Fri, 2005-10-28 at 03:31 +0100, Alistair John Strachan wrote:
> > > On Friday 28 October 2005 00:18, Claudio Scordino wrote:
> > > > Hi,
> > > >
> > > >     during the last years there has been a lot of discussion about
> > > > the "best" value of HZ... On i386 was 100, then became 1000, and
> > > > finally was set to 250. I'm thinking to do an evaluation of this
> > > > parameter using different architectures.
> > > >
> > > > Has anybody thought to give the possibility to modify the value of HZ
> > > > at boot time instead of at compile time ? This would allow to easily
> > > > test different values on different machines and create a table
> > > > containing the "best" value for each architecture...  At this moment,
> > > > instead, we have to recompile the kernel for each different value :(
> > > >
> > > > Do you think there would be much work to do that ?
> > > > Do you think it would be a desired feature the knowledge of the best
> > > > value for each architecture with more precision ?
> > >
> > > Google for "dynticks". There's obviously an overhead associated with HZ
> > > not being a constant (the compiler cannot optimise many expressions),
> > > but the feature is being worked on nonetheless.
> >
> > Well Linus had the best idea in that thread (as usual) which was to
> > implement "dynamic ticks" by leaving HZ a constant, setting it to a high
> > value, and skipping ticks when idle.  Has there been any work in that
> > direction?
>
> i did a bit of work in that area, but the stuff I came up with never
> seemed to work right, so I dropped it.

It's all still in development at the moment but not far from being available 
again. We stood back a bit to make some structural changes before trying to 
make it ready for prime time.

Cheers,
Con

Re: The "best" value of HZ

[Adrian Bunk]

On Fri, Oct 28, 2005 at 01:18:41AM +0200, Claudio Scordino wrote:

> Hi,

Hi Claudio,

>     during the last years there has been a lot of discussion about the "best" 
> value of HZ... On i386 was 100, then became 1000, and finally was set to 250.
> I'm thinking to do an evaluation of this parameter using different 
> architectures.
> 
> Has anybody thought to give the possibility to modify the value of HZ at boot 
> time instead of at compile time ? This would allow to easily test different 
> values on different machines and create a table containing the "best" value 
> for each architecture...  At this moment, instead, we have to recompile the 
> kernel for each different value :(
> 
> Do you think there would be much work to do that ? 
> Do you think it would be a desired feature the knowledge of the best value for 
> each architecture with more precision ?

the best value for HZ is not architecture specific, it depends on the 
usage pattern.

The rule is roughly:
- low HZ for computations
- high HZ for interactive usage

Making HZ selectable at boot time wouldn't be hard, but I doubt it's 
worth it because it would make the kernel both bigger and slower.

> Thanks,
> 
>       Claudio

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed