Better performance with 2.6

Better performance with 2.6

[albhaf]

I know that 2.6 has the ability to get better
performance than the other versions.
But what parts of the kernel has provided to this
performance upgrade?

/albhaf

Re: Better performance with 2.6

[Mike Fedyk]

albhaf wrote:
> I know that 2.6 has the ability to get better
> performance than the other versions.
> But what parts of the kernel has provided to this
> performance upgrade?

Preempt
BIO
Enhanced Locking

There are others I can't think of now, but these are definately high on 
the list.

Mike

Re: Better performance with 2.6

[Richard B. Johnson]

On Tue, 2 Mar 2004, Albert Hafvenstrom wrote:

> One of the things I found was that 2.6 detected my full CPU-capacity
> (nad even a bit more).
> Before, with 2.4, it showed my AMD Duron as 799.xxxx but now it is 800.047
> Is it because of some specific reason or does it just happen?
>
> /albhaf

Are you talking about BogoMips??  This is just how many twinkies
you can eat in a second with the current coding style in the
short timer counter. It has absolutely, positively, nothing to
do with "CPU capacity".

Cheers,
Dick Johnson
Penguin : Linux version 2.4.24 on an i686 machine (797.90 BogoMips).
            Note 96.31% of all statistics are fiction.

Re: Better performance with 2.6

[Chris Meadors]

On Tue, 2004-03-02 at 14:10, Richard B. Johnson wrote:

> Are you talking about BogoMips??  This is just how many twinkies
> you can eat in a second with the current coding style in the
> short timer counter. It has absolutely, positively, nothing to
> do with "CPU capacity".

He probably meant MHz.  But the same thing.  What difference does a
tenth of a MHz matter?

I do have a question about BogoMIPS.  I know they don't mean anything,
but why on my Opteron system with two processors that read the same on
the cpu MHz line, do my bogomips vary so much?

processor       : 0
cpu MHz         : 1393.980
bogomips        : 2736.12

processor       : 1
cpu MHz         : 1393.980
bogomips        : 3145.72


-- 
Chris

Re: Better performance with 2.6

[Albert Hafvenstrom]

One of the things I found was that 2.6 detected my full CPU-capacity (nad even a bit more).
Before, with 2.4, it showed my AMD Duron as 799.xxxx but now it is 800.047
Is it because of some specific reason or does it just happen?

/albhaf

Re: Better performance with 2.6

[Richard B. Johnson]

On Tue, 2 Mar 2004, Chris Meadors wrote:

> On Tue, 2004-03-02 at 14:10, Richard B. Johnson wrote:
>
> > Are you talking about BogoMips??  This is just how many twinkies
> > you can eat in a second with the current coding style in the
> > short timer counter. It has absolutely, positively, nothing to
> > do with "CPU capacity".
>
> He probably meant MHz.  But the same thing.  What difference does a
> tenth of a MHz matter?
>
> I do have a question about BogoMIPS.  I know they don't mean anything,
> but why on my Opteron system with two processors that read the same on
> the cpu MHz line, do my bogomips vary so much?
>
> processor       : 0
> cpu MHz         : 1393.980
> bogomips        : 2736.12
>
> processor       : 1
> cpu MHz         : 1393.980
> bogomips        : 3145.72
>

Because the loop-counter is called at different times, therefore
the cache has different stuff in it.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.24 on an i686 machine (797.90 BogoMips).
            Note 96.31% of all statistics are fiction.

Re: Better performance with 2.6

[Darren Williams]

Hi albhaf

Look at the slab allocator, this is a cache
for commonly used objects in the kernel.

For more info see the original document:
The Slab Allocator:
An object caching kernel memory allocator
Jeff Bonwick

and

mm/slab.c

A quick google should turn up the correct paper.

Darren

On Tue, 02 Mar 2004, albhaf wrote:

> I know that 2.6 has the ability to get better
> performance than the other versions.
> But what parts of the kernel has provided to this
> performance upgrade?
> 
> /albhaf
> 
> -
> 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/
--------------------------------------------------
Darren Williams <dsw AT gelato.unsw.edu.au>
Gelato@UNSW <www.gelato.unsw.edu.au>
--------------------------------------------------

Re: Better performance with 2.6

[Andrea Arcangeli]

On Tue, Mar 02, 2004 at 10:40:19AM -0800, Mike Fedyk wrote:
> albhaf wrote:
> >I know that 2.6 has the ability to get better
> >performance than the other versions.
> >But what parts of the kernel has provided to this
> >performance upgrade?
> 
> Preempt

preempt hurts a bit performance (at least when running a single threaded
app) with the object of helping interactivity, I say in theory because
a 2.4-aa based kernel has a worst case RT scheduler latency lower than
2.6 with preempt enabled (during I/O and some other workload).

Re: Better performance with 2.6

[Mike Fedyk]

Darren Williams wrote:
> Hi albhaf
> 
> Look at the slab allocator, this is a cache
> for commonly used objects in the kernel.
> 
> For more info see the original document:
> The Slab Allocator:
> An object caching kernel memory allocator
> Jeff Bonwick

Slab is in the 2.4 kernel also...