Uwe Klein wrote:
> Kevin Oberman wrote:
>> To get an idea of the fanaticism involved, several years ago, Kirk
>> McKusick and my former boss here in Berkeley counted the machine cycles
>> in the FreeBSD kernel for the PPS response(all done in the interrupt
>> service routine) and used that to correct the time. Of course,in the
>> days of 16 and 40 MHz systems, this was a bit more important.
> 
> Determinism has been lost.
> 
> Cache hits/misses, abitrated bus accesses, multicore contention, ...
> All have cost determinism in latency. You can count cycles
> but wont see the same count ever.

Which is one of the reasons the short pipeline and in-order Pentium made 
some pretty nice ntp servers.

This is coming back you know, in the form of Atom and Larrabee cpus: 
Many in-order cores with much more deterministic performance.

Terje

Terje

-- 
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"

_______________________________________________
questions mailing list
[email protected]
https://lists.ntp.org/mailman/listinfo/questions

Reply via email to