Re: Spinlocks: Factor our GENERIC_LOCKBREAK in order to avoid spinwith irqs disable

[Jeremy Fitzhardinge]

Nick Piggin wrote:
> Although, you wouldn't need to oversubscribe physical CPUs to hit
> suboptimal behaviour.
>
> Basically, I just ask for performance improvement to be measured
> with some "realistic" configuration, then it should be easier to
> justify.

Overcommitting cpus system-wide is pretty common.  A use-case is 
consolidating a pile of underused non-performance-critical servers into 
one physical machine with a lot of VMs.  It doesn't take much load to 
cause a dhcp request to the dhcp server to steal a cpu away from the 
mysql server while it holds a lock...

I'll mail out my patches so we can have a more concrete discussion shortly.

   J
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/