Been in the tree for ages and more importantly uses pidfiles. Once it's stable then baselayout and irqbalance will be bumped so that irqbalance is removed from baselayout and irqbalance gets its own init script. Thanks
how should we test this?
Good question. I have no idea really, but it's only useful for 2.4 kernels and 2.6 has it's own internal IRQ balancer. As I have no SMP machines I cannot test it :) BTW, I've committed -r1 versions of 0.08 and 0.12 with the init script in question (one with pidfile support, one without) So mark -r1 stable- thanks
(In reply to comment #2) > Good question. > > I have no idea really, but it's only useful for 2.4 kernels and 2.6 has it's > own internal IRQ balancer. amd64 doesn't support <2.6, please readd us if I'm missing something.
(In reply to comment #3) > (In reply to comment #2) > > Good question. > > > > I have no idea really, but it's only useful for 2.4 kernels and 2.6 has it's > > own internal IRQ balancer. > > amd64 doesn't support <2.6, please readd us if I'm missing something. OK, either stable 0.12-r1 or drop your amd64 keyword - your choice :)
(In reply to comment #4) > OK, either stable 0.12-r1 or drop your amd64 keyword - your choice :) > checked it out, was added by lv in 5/04. I can only guess it was needed back then. blubb will drop the keyword for us. Thanks for the heads up. Feel free to readd if I'm still missing something...
(In reply to comment #2) > I have no idea really, but it's only useful for 2.4 kernels and 2.6 has it's > own internal IRQ balancer. Sorry, but this is quite untrue since the userspace irqbalance is much better and preferred see this thread: http://lkml.org/lkml/2006/7/31/216
> Sorry, but this is quite untrue since the userspace irqbalance is much better > and preferred see this thread: > http://lkml.org/lkml/2006/7/31/216 > That was my understanding too....I have dozens of amd64 gentoo boxes deployed, and all are configured to use irqbalance for performance. In fact, I think redhat/et all have irqbalance enabled for amd64 by default. If we're wrong or the info is out of date, I'd definitly want to know.
ppc stable
(In reply to comment #6) > (In reply to comment #2) > > I have no idea really, but it's only useful for 2.4 kernels and 2.6 has it's > > own internal IRQ balancer. > Sorry, but this is quite untrue since the userspace irqbalance is much better > and preferred see this thread: > http://lkml.org/lkml/2006/7/31/216 > I was going by what someone (I forget who) told me in IRC. Looks like you're right.
Can we please have irqbalance unmasked again for amd64? Contrary to what was reported here earlier, amd64 kernels do not do internal irqbalancing. Or at least the current stable gentoo-sources, 2.6.17-r8 does not. Judging by the kernel sources, in-kernel irq balancing is a feature specific to i386.
(In reply to comment #10) > Can we please have irqbalance unmasked again for amd64? Contrary to what was > reported here earlier, amd64 kernels do not do internal irqbalancing. Or at > least the current stable gentoo-sources, 2.6.17-r8 does not. Judging by the > kernel sources, in-kernel irq balancing is a feature specific to i386. amd64 team, looks like your keyword is needed again :P
x86 is stable ^.^
Thanks for the correction, it is marked stable now on amd64 too.
How about adding this as a base layout requirement? Fedora and others ship with it. With the rise of dual core+ systems, this seems like a necessary package. Judging by the confusion _developers_ had on its obsolescence and effect, imagine what the users will think :O? http://lwn.net/Articles/213731/ "This new version knows about, and optimizes for, Dual and Quad core, and knows about MSI, PCI-Express, NAPI, Cache domains, processor sockets etc etc. In addition, the new irqbalance switches to a power-save mode when there is little irq load on the system, trying to preserve power by avoiding waking up processors more than needed." So there is benefit to mobile systems as well.
(In reply to comment #14) > How about adding this as a base layout requirement? No. baselayout supports FreeBSD now, where this would not obviously work. Also, SMP systems work perfectly happily without this and it has it's own init script, therefore it's 100% optional. iproute2 isn't a requirement for similar reasons even though it's technically superior to ifconfig from net-tools.