Re: sched_clock() uses are broken

[Nick Piggin <nickpiggin@yahoo.com.au>]

Mike Galbraith wrote:

>Sorry for yet another reply, but running the old starvation testcase
>that caused sched_clock() to be born in the first place tickled my
>funny-bone.  With that running and hitting 300k context switches...
>
>now: 2100508962835 tick: 2100508972067 stamp: 2100508961220 total: 2906
>now: 2101531243883 tick: 2101531251877 stamp: 2101531238543 total: 2924
>now: 2102695422392 tick: 2102695431699 stamp: 2102695418265 total: 2940
>
>Accounting?  Not :)
>

Yeah I agree with Andi that this accounting stuff is probably done
for some POSIX conformance that doesn't matter. Actually it is worse
than that because if anyone _really_ did need it, then they'll get a
horrible surprise when their system mysteriously fails in production.

It should either get ripped out, or perhaps converted to use jiffies
until a sane high resolution, low overhead scheme is developed (if
ever). And that would exclude something that does this accounting in
fastpaths for the 99.99% of processes that never use it.

--

Send instant messages to your online friends http://au.messenger.yahoo.com