21 Jul
2011
21 Jul
'11
1:20 a.m.
Excerpts from David Olofson's message of 2011-07-21 01:03:32 +0200:
On Thursday 21 July 2011, at 00.36.55, Philipp
Überbacher
<hollunder(a)lavabit.com> wrote:
[...]
I do wonder whether the occasional peaks are due to the mainboards or
the CPUs. At least in the wild mainboards seem to have a lot of
influence (hardware interrupts or however they're called).
> ..and this latency plot is stunning:
> https://www.osadl.org/Latency-plot-of-system-in-rack-4-slot.qa-latencyplo
> t-r4s6.0.html?latencies=&showno=&slider=57
[...]
The plot really does look stunning, strangly (?)
not on other machines.
https://www.osadl.org/Latency-plot-of-system-in-rack-4-slot.qa-latencyplot-
r4s7.0.html
https://www.osadl.org/Latency-plot-of-system-in-rack-4-slot.qa-latencyplot
-r4s8.0.html
No idea what those plots tell about real world usage. It's good to get
another set of patches though.
In terms of worst case figures, these plots look a bit like what I've seen
with RT-Linux and RTAI on various hardware (PII/III workstations via Geode
SBCs through Intel Core based Celerons on industrial Mini-ITX boards), though
with "true" RT kernels, one tends to get a lot of very low latency points, and
only the occasional peak.
SBCs with lowpower CPUs (Geode and the like) tend to perform a lot worse than
"proper" laptop and desktop CPUs. Memory and/or cache bandwidth issues, maybe?