[MAGEEC] Energy cost of context switching
simon at compsci.bristol.ac.uk
Wed Nov 20 13:09:25 GMT 2013
I recall that we discussed this idea in an ENTRA meeting, with the main feeling being that event waiting would be the.most efficient implementation, but that it often doesn't occur due to lack of visibility of the cost of polling.
It would be a nice study to see the difference when waiting on an input, but I feel that there must be a paper on the energy needs and the scheduling impacts somewhere. Perhaps a little digging to do first.
Finally, yes, context switching is getting cheaper with the more modern ARMs, but cheaper means faster and an interesting observation is that the programmer gets less choice about the context that is preserved. This means that sometimes too much contrxt is stored to main memory, which will burn additional energy. This.may be worth looking at too.
-------- Original message --------
From: Joern Rennecke <joern.rennecke at embecosm.com>
To: Jeremy Bennett <jeremy.bennett at embecosm.com>,MAchine Guided Energy Efficient Compilation <mageec at mageec.org>
Subject: Re: [MAGEEC] Energy cost of context switching
On 19 November 2013 18:54, Jeremy Bennett <jeremy.bennett at embecosm.com> wrote:
> On 19/11/13 18:51, Kerstin Eder wrote:
>> Hi Jeremy,
>> This would make an interesting student project - investigate polling vs
>> interrupts for different platforms / loads. ;)
I think the most potential is where you change the system design.
I.e. instead of one high-performance universal processor, you
have a number of low-power special-purpose processors.
mageec mailing list
mageec at mageec.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the mageec