[ovs-dev] [PATCH v2] ovn-controller: eliminate stall in ofctrl state machine

Ben Pfaff blp at ovn.org
Mon Jul 25 00:52:18 UTC 2016


On Sun, Jul 24, 2016 at 04:44:12PM -0500, Ryan Moats wrote:
> Ben Pfaff <blp at ovn.org> wrote on 07/24/2016 03:17:18 PM:
> 
> > From: Ben Pfaff <blp at ovn.org>
> > To: Ryan Moats/Omaha/IBM at IBMUS
> > Cc: Lance Richardson <lrichard at redhat.com>, dev at openvswitch.org
> > Date: 07/24/2016 03:17 PM
> > Subject: Re: [ovs-dev] [PATCH v2] ovn-controller: eliminate stall in
> > ofctrl state machine
> >
> > On Sat, Jul 23, 2016 at 09:35:38AM -0500, Ryan Moats wrote:
> > > While this code is cleaner and more robust, running it against the same
> > > end-to-end tests that I applied to Lance's original/V2 patches doesn't
> > > result in the same small performance gain at larger port densities
> > > and reveals a very slight performance penalty for lower port densities.
> > >
> > > This wasn't what I was expecting and leaves me wondering what is
> > > causing the difference.
> >
> > Are your performance differences within the normal margin of variation?
> 
> Let me put it this way:
> 
> At the higher port density (50 ports/switch) I was testing, I am no
> longer 95% confident of a performance gain, while at the lower port
> density (10 ports/switch), I am now 95% confident of a performance
> hit.
> 
> Now, I suspect that if I go up to even higher port densities I may
> still see an improvement, but I haven't tested that.  Further, I
> *think* the reason for this change is that Lance's patch processed
> more messages in a go and so had a chance of starving out other
> parts of the code, while yours most definitely doesn't...
> 
> Thus, I'm not asking for a revert or a redesign, I'm just reporting what
> I saw and wondering...

OK, thanks.  I interpreted this as reporting a bug but it's more nuanced
than that.



More information about the dev mailing list