IRQ Balance Question for Single but Multi-Core Processors

IRQ Balance Question for Single but Multi-Core Processors

[Justin Piszcz]

Question regarding the IRQ balance daemon and the 2.6.x kernel.

For a single-processor but dual or quad core CPU, should one be running 
the IRQ balancing daemon, will it result in increased performance?

Or is IRQ balance mainly targeted at physical multi-processor systems?

Justin.

Re: IRQ Balance Question for Single but Multi-Core Processors

[Jan Engelhardt]

On Jun 24 2007 16:29, Justin Piszcz wrote:
>
> Question regarding the IRQ balance daemon and the 2.6.x kernel.
>
> For a single-processor but dual or quad core CPU, should one be running the IRQ
> balancing daemon, will it result in increased performance?
>
> Or is IRQ balance mainly targeted at physical multi-processor systems?

There must be a reason it exists :)
For MC or MP or MCMP systems, it should be helpful indeed.
HT however, not so sure. But then again, MCHT and MCMPHT maybe.

(multi-core, multi-processor, hyperthreading, and combinations thereof.)


	Jan
-- 

Re: IRQ Balance Question for Single but Multi-Core Processors

[Arjan van de Ven]

Justin Piszcz wrote:
> Question regarding the IRQ balance daemon and the 2.6.x kernel.
> 
> For a single-processor but dual or quad core CPU, should one be running 
> the IRQ balancing daemon, will it result in increased performance?

it's a tradeoff.

for cores/threads that share cache, there's no point to run a daemon 
all the time, they share so many resources that things don't really 
matter.....

irqbalance will, as a result, set up a one time static mapping and 
then exit on such systems; this will give a "reasonable" spread of 
interrupts, but won't cost you any cpu after that.

On "real" smp systems, the performance and power tradeoffs are 
different and there irqbalance will keep an eye on things over time...