Re: SMP

[Phil Brutsche <pbrutsch@creighton.edu>]

A long time ago, in a galaxy far, far way, someone said...

> Phil Brutsche wrote:
> > Expect few to no problems if you upgrade to 2.2.x as soon as possible (SMP
> > support in 2.0.x, which is the default in slink, isn't all that hot).

> What do you mean by that? I have a default Slink installation, I
> recompiled the kernel for SMP, and it _seems_ to work correctly (e.g.
> MATLAB benchmark shows a significant improvement of performance). In
> fact I'm a beginner in SMP so should I expect any problems in the
> future?

I'm not saying that 2.0 SMP doesn't work, 'cause it does.  It's just that
2.2 SMP (and 2.2 in general, IMO) works better.  Most of the difference
comes down to technical terms; specifically "2.0 used a single kernel
lock, while 2.2 has much more fine-grained locking, and 2.3 has kernel
locking even more fine-grained than 2.2".

Having a fine-grained kernel lock has the benefit of higher scalability
and performance on multi processor systems, but hurts if if you have
"only" a single processor system.  It's actually a bit more complex than
that, but that's the gist of it.

SMP is 2.2 is also more mature; many poeple have noticed stability
improvements under high load going from 2.0 SMP to 2.2 SMP.  2.3 SMP is
proving to be even better in terms of performance and stability.

There's all sorts of other benefits to 2.2.x that make it a worthwhile
upgrade.

> BTW: how to see how my system distributes the workload between the two
> processors?

top, I've seen, shows you something like that.

-- 
----------------------------------------------------------------------
Phil Brutsche					pbrutsch@creighton.edu

"There are two things that are infinite; Human stupidity and the
universe. And I'm not sure about the universe." - Albert Einstein