Re: Router take advantage of device polling

2004-05-27 Thread T. Srikanth
Jeffrey,
Yes, there is. At a certain level of traffic, network interrupts
happen all too often to justify the processing overhead associated
with the interrupt service routines, and hence the network activity
is best polled. Is 15Mbps the level for enabling polling? Well,
this traffic level depends on the underlying hardware capabilities
as well. You decide.
--Srikanth
Jeffrey Lyon wrote:
I am currently running a router pushing about 15 megabit of traffic,
with CPU usage like so:
root 25  5.9  0.0 0   12  ??  WL   10:16AM  47:49.70  (irq7: bge0 amr0)
root 14  5.9  0.0 0   12  ??  WL   10:16AM  55:15.63  (swi1: net)
root 26  3.0  0.0 0   12  ??  LL   10:16AM  28:42.81  (irq5: bge1)
I am told that activating the device_polling kernel option will cause a 
substantial reduction in this usage. Is there any truth to that statement?

Thanks.
--
Jeffrey Lyon, Independent Associate
jeffreylyon at prepaidlegal.com 
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
http://www.prepaidlegal.com/hub/jeffreylyon
Pre-Paid Legal Services, Inc.
___
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]


Re: Router take advantage of device polling

2004-05-22 Thread Matthew Seaman
On Sat, May 22, 2004 at 06:51:32AM -0400, Jeffrey Lyon wrote:
 I am currently running a router pushing about 15 megabit of traffic, 
 with CPU usage like so:
 
 root 25  5.9  0.0 0   12  ??  WL   10:16AM  47:49.70  (irq7: 
 bge0 amr0)
 root 14  5.9  0.0 0   12  ??  WL   10:16AM  55:15.63  (swi1: net)
 root 26  3.0  0.0 0   12  ??  LL   10:16AM  28:42.81  (irq5: bge1)
 
 I am told that activating the device_polling kernel option will cause a 
 substantial reduction in this usage. Is there any truth to that statement?

Probably not.  I see you are using bge(4) devices -- there's nothing
in the polling(4) or bge(4) man pages to suggest that the bge
interface does support polling, and the code in /usr/src/sys/dev/bge
makes no reference to the DEVICE_POLLING cpp macro, so it's unlikely
to help with that hardware.

Assuming you've got suitable hardware, whether device polling gains
you much, or anything at all depends on the nature of the traffic
you're dealing with.  There are about dozen kernel tunables that you
can use to optimize traffic flows. See the polling(4) man page for
more information.

Cheers,

Matthew


-- 
Dr Matthew J Seaman MA, D.Phil.   26 The Paddocks
  Savill Way
PGP: http://www.infracaninophile.co.uk/pgpkey Marlow
Tel: +44 1628 476614  Bucks., SL7 1TH UK


pgpUU7kuDeBRI.pgp
Description: PGP signature