4K vs 8K stacks- Which to use?

4K vs 8K stacks- Which to use?

[Justin Piszcz]

I use an array of machines with all sorts of CPU's (but no 64bit CPU's 
yet):

Which should I use for each CPU?
Which is better and why?

Pentium 1 CPU's
Cyrix P150 (120MHZ)
Pentium 2 CPU's
Pentium 3 CPU's
Pentium 4 W/HT
Pentium 4 W/OUT HT

Re: 4K vs 8K stacks- Which to use?

[Ikke]

Hi,

I got a P2 here, and got some problems that *could* be related to 4k stacks.
I'm running 2.6.7-redeeman3 now (i.e. 2.6.7-mm3+reiser4+nick's sheduler).
If I configure it using 4k stacks and APM as module (not loaded) I get
crashes (Page Faults I think). Same configuration, but 4k stacks
disabled and APM module disabled, runs stable.

Just for your interest ;)

Regards, Ikke

On Sat, 3 Jul 2004 06:50:58 -0400 (EDT), Justin Piszcz
<jpiszcz@lucidpixels.com> wrote:
> I use an array of machines with all sorts of CPU's (but no 64bit CPU's
> yet):
> 
> Which should I use for each CPU?
> Which is better and why?
> 
> Pentium 1 CPU's
> Cyrix P150 (120MHZ)
> Pentium 2 CPU's
> Pentium 3 CPU's
> Pentium 4 W/HT
> Pentium 4 W/OUT HT
> 
> -
> 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: 4K vs 8K stacks- Which to use?

[Chris Wedgwood]

On Sun, Jul 04, 2004 at 12:55:07PM +0200, Ikke wrote:

> I got a P2 here, and got some problems that *could* be
> related to 4k stacks.  I'm running 2.6.7-redeeman3 now
> (i.e. 2.6.7-mm3+reiser4+nick's sheduler).  If I configure
> it using 4k stacks and APM as module (not loaded) I get
> crashes (Page Faults I think). Same configuration, but 4k
> stacks disabled and APM module disabled, runs stable.

Try turning on the stack overflow checking and see what that
spits out.



  --cw

Re: 4K vs 8K stacks- Which to use?

[Ikke]

I will when I get my next kernel (compiling one takes some time here).

On Sun, 4 Jul 2004 05:34:31 -0700, Chris Wedgwood <cw@f00f.org> wrote:
> On Sun, Jul 04, 2004 at 12:55:07PM +0200, Ikke wrote:
> 
> > I got a P2 here, and got some problems that *could* be
> > related to 4k stacks.  I'm running 2.6.7-redeeman3 now
> > (i.e. 2.6.7-mm3+reiser4+nick's sheduler).  If I configure
> > it using 4k stacks and APM as module (not loaded) I get
> > crashes (Page Faults I think). Same configuration, but 4k
> > stacks disabled and APM module disabled, runs stable.
> 
> Try turning on the stack overflow checking and see what that
> spits out.
> 
> 
>   --cw
>

Re: 4K vs 8K stacks- Which to use?

[Justin Piszcz]

> does it matter?
Well the option is available :)

> why do you think it would be processor-specific?
Well I know IA32 is limited to a 4096 byte page size in the Linux Kernel; 
hence filesystems can only use 4KB blocks on IA32, therefore I was 
wondering if anything might change in 64bit land?

  On Sat, 3 Jul 2004, Mark 
Hahn wrote:

>> I use an array of machines with all sorts of CPU's (but no 64bit CPU's
>> yet):
>>
>> Which should I use for each CPU?
>
> does it matter?  4k stacks are primarily for benchmark machines
> under high load, where smaller stacks mean more threads available,
> and less chance of failing to allocate two contiguous pages.
>
>> Which is better and why?
>
> can't possibly matter for any normal use.
>
>> Pentium 1 CPU's
>> Cyrix P150 (120MHZ)
>> Pentium 2 CPU's
>> Pentium 3 CPU's
>> Pentium 4 W/HT
>> Pentium 4 W/OUT HT
>
> why do you think it would be processor-specific?
>

Re: 4K vs 8K stacks- Which to use?

[William Lee Irwin III]

On Sat, 3 Jul 2004, Mark Hahn wrote:
>> why do you think it would be processor-specific?

On Sat, Jul 03, 2004 at 10:44:44AM -0400, Justin Piszcz wrote:
> Well I know IA32 is limited to a 4096 byte page size in the Linux Kernel; 
> hence filesystems can only use 4KB blocks on IA32, therefore I was 
> wondering if anything might change in 64bit land?

64-bit ports tend to have stacks around twice the native pagesize, e.g.
sparc64 has 16K stacks and 8K pages, similar to 32-bit ports, except for
ia64. The ia32 port has adopted the order 0 stacksize in the interest
of space savings relatively early among Linux ports. It would be nice
to get the same for all architectures, as higher-order allocations are
not good to have as anything but speculative (and obviously stacks are
not variable-sized) for reasons of fragmentation.


-- wli