Article at 2CPU.com - Linux 2.6 and Hyper-Threading

[Marc Schwartz]

Hi all,

Just as an FYI to the group, the following article is new at 2CPU.com:

Linux 2.6 and Hyper-Threading 
http://www.2cpu.com/articles/ht_linux/

The final paragraph:

"The final word on Linux and hyper-threading? While the majority of
benchmarks we've looked at here today have shown improvements, it's
still important that you do your own testing with the applications you
use on a daily basis. On the surface, HT certainly looks promising but
it's clear that there are still situations out there where HT is going
to negatively impact the performance of an application. Before rolling
out those new workstations or servers, do your own testing in
pre-production to find out what hyper-threading can do for you."


Enjoy,

Marc Schwartz

[Dave Jones]

On Wed, 2004-02-25 at 17:23, Marc Schwartz wrote:

>  Linux 2.6 and Hyper-Threading 
>  http://www.2cpu.com/articles/ht_linux/
>  
>  The final paragraph:
>  
>  "The final word on Linux and hyper-threading? While the majority of
>  benchmarks we've looked at here today have shown improvements, it's
>  still important that you do your own testing with the applications you
>  use on a daily basis. On the surface, HT certainly looks promising but
>  it's clear that there are still situations out there where HT is going
>  to negatively impact the performance of an application. Before rolling
>  out those new workstations or servers, do your own testing in
>  pre-production to find out what hyper-threading can do for you." 
Very sensible.  For a lot of folks, it makes very little improvement
whatsoever. Personally, my most common workload (compiling kernels)
wins quite considerably with hyperthreading (last time I measured
it took over a minute off the build).

It's worth pointing out that at this point, the scheduler in the 2.6
kernel is still lacking some HT enhancements.

	Dave

[Marc Schwartz]

On Wed, 2004-02-25 at 11:32, Dave Jones wrote:

>  It's worth pointing out that at this point, the scheduler in the
> 2.6
>  kernel is still lacking some HT enhancements. 

Is that something that will come in time to 2.6 or is that significant
enough to defer to 2.8?

Thanks Dave.

Marc

[Dave Jones]

On Wed, 2004-02-25 at 17:43, Marc Schwartz wrote:

>  > It's worth pointing out that at this point, the scheduler in
> the 2.6
>  > kernel is still lacking some HT enhancements.
>  Is that something that will come in time to 2.6 or is that significant
>  enough to defer to 2.8? 
It'll get fixed for 2.6. Folks are kicking around patches already,
and some of them are already in 2.6-mm which is a promising start.

	Dave

[Marc Schwartz]

On Wed, 2004-02-25 at 11:56, Dave Jones wrote:
>  On Wed, 2004-02-25 at 17:43, Marc Schwartz wrote:
>  
>  > > It's worth pointing out that at this point, the scheduler in the 2.6
>  > > kernel is still lacking some HT enhancements.
>  > Is that something that will come in time to 2.6 or is that significant
>  > enough to defer to 2.8?
>  
>  It'll get fixed for 2.6. Folks are kicking around patches already,
>  and some of them are already in 2.6-mm which is a promising start.
>  
>  	Dave 

Great.  Thanks Dave.

Marc