[MAGEEC] Energy cost of context switching

[Jeremy Bennett]

Chris Shaw of ARM had a slide at the NMI meeting yesterday, in which he
suggested you should always prefer interrupts over polling.

I challenged this (as I often do), since it actually depends on your
latency and throughput, but in many circumstances polling, or at least
time-triggered polling is going to be better.

The assumption I was making was that context switching uses a lot of
energy. Chris Shaw is pushing back on this, particularly for the
Cortex-M series, which have very efficient context switching.

Do we have any data on this. If not how easy would it be to create some
artificial loads and compare polling with switching in different scenarios?


Jeremy

-- 
Tel:      +44 (1590) 610184
Cell:     +44 (7970) 676050
SkypeID: jeremybennett
Twitter: @jeremypbennett
Email:   jeremy.bennett at embecosm.com
Web:     www.embecosm.com