Re: Linux v3.12 + pv ticketlocks

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Sep 11, 2013 at 09:05:05PM +0200, Sander Eikelenboom wrote:
> 
> Wednesday, September 11, 2013, 8:41:29 PM, you wrote:
> 
> > On Wed, Sep 11, 2013 at 08:25:01PM +0200, Sander Eikelenboom wrote:
> >> Hi Konrad,
> >> 
> >> Just to confirm:
> >> 
> >> - PV ticketlocks is on by default and doesn't neet a config option, to disable it at boot time it's  possible to use "xen_nopvspin" as a kernel commandline parameter
> 
> > Yes.
> >> - CONFIG_PARAVIRT_SPINLOCKS should be off, since that would enable the old bytelock pv-spinlock implementation which could now probably be removed ?
> 
> > You still need CONFIG_PARAVIRT_SPINLOCKS=y to take advantage of it.
> 
> > The bytelock pvspinlock is gone. If you compile without CONFIG_PARAVIRT_SPINLOCKS
> > then you will only use ticketlocks for all types of guests and baremtal. No
> > 'PV' variant.
> 
> > The CONFIG_PARAVIRT_SPINLOCKS=y enables the PV ticketlock for all guests.
> 
> Ok so the KConfig (help) entry for CONFIG_PARAVIRT_SPINLOCKS should perhaps also be updated ?

Yes!
> 
> Since it for example still contains the "Unfortunately the downside is an up to 5%
> performance hit on native kernels, with various workloads." which afaik was one the things the PV ticketlock migitated,
> so it would be more feasible for distributions to have this option on for all general distro kernels ?

Yikes. That is true - needs to be updated as the results were the opposite:

 Results:
    =======
    setup: 32 core machine with 32 vcpu KVM guest (HT off)  with 8GB RAM
    base = 3.11-rc
    patched = base + pvspinlock V12
    
    +-----------------+----------------+--------+
     dbench (Throughput in MB/sec. Higher is better)
    +-----------------+----------------+--------+
    |   base (stdev %)|patched(stdev%) | %gain  |
    +-----------------+----------------+--------+
    | 15035.3   (0.3) |15150.0   (0.6) |   0.8  |
    |  1470.0   (2.2) | 1713.7   (1.9) |  16.6  |
    |   848.6   (4.3) |  967.8   (4.3) |  14.0  |
    |   652.9   (3.5) |  685.3   (3.7) |   5.0  |
    +-----------------+----------------+--------+

(I don't have the baremetal one - but it was pretty much 0%).

Please if you get a chance - please do send a patch to x86 folks
altering this.

Thanks!