interrupt numbers
Matthew Dillon
dillon at apollo.backplane.com
Thu Apr 13 10:13:03 PDT 2006
:
:Terry Tree <terry.tree at xxxxxxxxx> wrote:
: > I take it everyone else is just as lost as me on how answer this one ?
:
:Answer what? I didn't see any question in that message. :-)
:(Apart from that, it doesn't seem to be related to DragonFly BSD.)
:
:Best regards
: Oliver
:
:--
:Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing
I didn't see a question there either. I did look at the web site,
but while the graphs look quite interesting (though FreeBSD related,
not DragonFly related), a vertical scale of 'ticks' rather than
nanoseconds or microseconds doesn't really help one get any sense of
scale.
In anycase, apart from the well understood overheads of issuing any
sort of 'read' operation on an I/O device, I'm not sure how applicable
any of that would be to a high performance system design since
all the interrupts we care about (network, disk) are able to present
multiple events per interrupt.
I ran some interrupt overhead tests for a Bay Lisa talk a few months
ago, which you can find here:
http://www.dragonflybsd.org/docs/LISA200512/index.html
That particular machine was operating in PIC mode (because Shuttle's
BIOS's ACPI and MP interrupt tables are completely broken). Interrupt
overhead was approximately 1uS. I am not particularly concerned
over 1uS of interrupt overhead.
It will become even less relevent as both Intel and AMD shift more of
their production into multi-core cpus.
-Matt
Matthew Dillon
<dillon at xxxxxxxxxxxxx>
More information about the Kernel
mailing list