Re: [PATCH RFC V6 0/11] Paravirtualized ticketlocks

From: Raghavendra K T
Date: Wed Mar 28 2012 - 14:21:42 EST


On 03/28/2012 09:39 PM, Alan Meadows wrote:
I am happy to see this issue receiving some attention and second the
wish to see these patches be considered for further review and inclusion
in an upcoming release.

Overcommit is not as common in enterprise and single-tenant virtualized
environments as it is in multi-tenant environments, and frankly we have
been suffering.

We have been running an early copy of these patches in our lab and in a
small production node sample set both on3.2.0-rc4 and 3.3.0-rc6 for over
two weeks now with great success. With the heavy level of vCPU:pCPU
overcommit required for our situation, the patches are increasing
performance by an _order of magnitude_ on our E5645 and E5620 systems.


Thanks Alan for the support. I feel timing of this patch was little bad
though. (merge window)



Looks like a good baseline on which to build the KVM
implementation. We
might need some handshake to prevent interference on the host
side with
the PLE code.


I think I still missed some point in Avi's comment. I agree that PLE
may be interfering with above patches (resulting in less performance
advantages). but we have not seen performance degradation with the
patches in earlier benchmarks. [ theoretically since patch has very
slight advantage over PLE that atleast it knows who should run next ].

So TODO in my list on this is:
1. More analysis of performance on PLE mc.
2. Seeing how to implement handshake to increase performance (if PLE +
patch combination have slight negative effect).

Sorry that, I could not do more analysis on PLE (as promised last time)
because of machine availability.

I 'll do some work on this and comeback. But in the meantime, I do not
see it as blocking for next merge window.


Avi, Thanks for reviewing. True, it is sort of equivalent to PLE on
non PLE machine.

Ingo, Peter,
Can you please let us know if this series can be considered for next
merge window?
OR do you still have some concerns that needs addressing.

I shall rebase patches to 3.3 and resend. (main difference would be
UNINLINE_SPIN_UNLOCK and jump label changes to use
static_key_true/false() usage instead of static_branch.)

--
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/