[Cbe-oss-dev] Required cycles for an SPE context-switch
Michael Ellerman
michael at ellerman.id.au
Sat Jan 19 14:59:20 EST 2008
On Fri, 2008-01-18 at 17:08 +0100, csc81 at csc81.de wrote:
> Hello all :)
>
> For my diploma thesis I would like to know the average costs of an SPE
> context-switch on a PS3 running Fedorra Core 7 Linux (expressed in core
> clock frequency cylces or microseconds) in absence of contention and
> under load. A rough estimation would be sufficient. The costs should
> include the communication over the EIB and the computational costs of
> the linux kernel (SPE scheduler and maybe others).
>
> My goal is to create a diagram that shows the percentage which is needed
> for context switching on N SPEs depending on different SPE Scheduler
> frequencies relative to the available resources on one Cell processor.
>
> I need this information to estimate the overhead of the preemptive
> programming model. In particular the worst case is interesting.
>
> Does anyone know a rough value or at least a paper or a documentation
> where I can find this information?
Section 12 of the CBE programming handbook describes the sequence:
http://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/9F820A5FFA3ECE8C8725716A0062585F
I don't know of any estimate of the cost, other than "lots". If you can
come up with a figure I think we'd all be very interested.
The best approach I think would be to instrument the kernel and/or your
spe app to measure the overhead.
cheers
--
Michael Ellerman
OzLabs, IBM Australia Development Lab
wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)
We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.ozlabs.org/pipermail/cbe-oss-dev/attachments/20080119/e37a84e3/attachment.pgp>
More information about the cbe-oss-dev
mailing list