cvs commit: src/sys/amd64/conf NOTES src/sys/i386/conf NOTES src/sys/pc98/conf NOTES

Gleb Smirnoff glebius at freebsd.org
Wed Feb 23 08:58:15 GMT 2005


On Wed, Feb 23, 2005 at 10:36:39AM +0200, Ruslan Ermilov wrote:
R> > Actually, the block with error can be safely removed from kern_poll.c before
R> > we polish our patches. There is no problem running SMP + polling. Probably,
R> > there were some problems in the past.
R> > 
R> > I think the block can be removed now and MFCed before 5.4-RELEASE. I'm quite
R> > sure that our patches for Giant-less polling will not be MFCed soon.
R> > 
R> Take a look at http://info.iet.unipi.it/~luigi/polling/, the last
R> paragraph on the page.  I believe it still holds true these days.

I completely agree here, but this is not a reason to deny such configuration.
What is nice in polling, is that some amount of CPU resources is reserved
for userland. So, if box is not doing a pure forwarding job, but is a web
server which is loaded both by userland applications and interrupts, a setup
with polling and SMP will be very appropriate.

-- 
Totus tuus, Glebius.
GLEBIUS-RIPN GLEB-RIPE


More information about the cvs-all mailing list