Date: Wed, 04 Mar 2009 01:50:13 -0700 (MST) From: "M. Warner Losh" <imp@bsdimp.com> To: hselasky@c2i.net Cc: freebsd-current@freebsd.org, freebsd-usb@freebsd.org Subject: Re: Knobs to reduce PCI interrupt latency Message-ID: <20090304.015013.-862235060.imp@bsdimp.com> In-Reply-To: <200903040942.39191.hselasky@c2i.net> References: <200903032243.31914.hselasky@c2i.net> <200903041028.37027.man@email.com.ua> <200903040942.39191.hselasky@c2i.net>
next in thread | previous in thread | raw e-mail | index | archive | help
In message: <200903040942.39191.hselasky@c2i.net>
Hans Petter Selasky <hselasky@c2i.net> writes:
: Do we have any knobs in FreeBSD to reduce the interrupt latency? I am
: currently seeing performance differences between 8.x and 7.x. Anyone have any
: ideas?
If we did, don't you think they would be enabled by default :)
Seriously, I measured interrupt latency on a 7.0-current system at
around 20us for a fast interrupt/filter (300MHz or 400MHz low-power
CPU). For fast machines, this can approach 1us, which is hard to
measure with the setup I had for the 300MHz case.
Deferring work to the scheduled interrupt can be useful, especailly if
you can keep the pipeline full such that the filter can kick off the
next set of transactions quickly, and then call the completion
routines on the last set in parallel in the ithread. This works well
for most random-access things, but less well for single threaded,
sequential benchmarks.
Warner
: --HPS
:
: On Wednesday 04 March 2009, Artyom Mirgorodsky wrote:
: > >Repeat the same test using FreeBSD -current.
: > >
: > >a) On the machine where it is slow.
: >
: > vmstat -i ; sleep 1 ; vmstat -i
: > interrupt total rate
: > irq1: atkbd0 233 2
: > irq14: ata0 85 0
: > irq16: vgapci0 5377 52
: > irq21: hdac0 ohci0 742 7
: > irq22: nfe0 ehci0 23610 229
: > irq23: atapci1 5405 52
: > cpu0: timer 203959 1980
: > cpu1: timer 200914 1950
: > Total 440325 4275
: > interrupt total rate
: > irq1: atkbd0 234 2
: > irq14: ata0 85 0
: > irq16: vgapci0 5439 52
: > irq21: hdac0 ohci0 742 7
: > irq22: nfe0 ehci0 24621 236
: > irq23: atapci1 5405 51
: > cpu0: timer 205981 1980
: > cpu1: timer 202937 1951
: > Total 445444 4283
: >
: > >I think the reduced performance can be explained by a clamp on the
: > > interrupt rate around 1000 interrupts per second instead of 8000. Maybe
: > > someone has an explanation for this?
: >
: > You right, the interrupt rate around 1000 (1011) on this machine, but on
: > FreeBSD 7.1 more 3000. If it is some kind of interrupt aggregation, may be
: > I can try to turn it off?
: >
: > >b) On the machine where it is fast.
: >
: > vmstat -i ; sleep 1 ; vmstat -i
: > interrupt total rate
: > irq4: uart0 4154 0
: > irq14: ata0 472922 0
: > irq15: ata1 26 0
: > irq18: em0 752711 0
: > irq21: ahc0 53 0
: > irq23: ehci0 103456 0
: > irq24: em1 147 0
: > cpu0: timer 1551216517 2000
: > cpu1: timer 1551195251 2000
: > Total 3103745237 4001
: > interrupt total rate
: > irq4: uart0 4154 0
: > irq14: ata0 472923 0
: > irq15: ata1 26 0
: > irq18: em0 752713 0
: > irq21: ahc0 53 0
: > irq23: ehci0 110949 0
: > irq24: em1 147 0
: > cpu0: timer 1551218551 2000
: > cpu1: timer 1551197285 2000
: > Total 3103756801 4001
:
:
: _______________________________________________
: freebsd-usb@freebsd.org mailing list
: http://lists.freebsd.org/mailman/listinfo/freebsd-usb
: To unsubscribe, send any mail to "freebsd-usb-unsubscribe@freebsd.org"
:
:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20090304.015013.-862235060.imp>