ULE and Prescott question

[Ivan Voras]

RW wrote:
> On Wed, 22 Jul 2009 22:33:46 +0200
> Ivan Voras <ivoras at freebsd.org> wrote:
> 
>> Scott Bennett wrote:
>>>      This is a curiousity question.  I'm running 7.2-STABLE at
>>> present on an old Inspiron XPS, which has a 3.4 GHz P4 Prescott
>>> CPU.  I have hyperthreading enabled in the kernel.  The question
>>> is:  is there any appreciable performance difference to be expected
>>> with this hardware setup between the ULE scheduler and the 4BSD
>>> scheduler?  Or does the fact that there is only one core eliminate
>>> any difference in performance characteristics?
>> I'd guess the second thing. It's not like there's cache to be shared
>> between cores, etc. 
> 
> But with hyperthreading enabled, don't you have virtual CPUs sharing
> L1 cache 

Yes,

> rather that cores sharing L2 cache, making the case for ULE
> even stronger?

If you're thinking about ULEs "soft-pinning" of processes to CPUs then I
don't think so for two reasons: it's not like 4BSD forces processes
ping-ponging everywhere - for 2 logical CPUs it's not that there's much
choice of where to schedule a process - and thread switches between HTT
logical CPUs is supposed to be cheap - I think since the L1 is shared,
HTT cores have access to cached data from "the other" core for no cost.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 259 bytes
Desc: OpenPGP digital signature
Url : http://lists.freebsd.org/pipermail/freebsd-questions/attachments/20090723/123dedf3/signature.pgp