On 21 July 2014 10:28, Joe Stringer wrote:
> On 19 July 2014 06:13, Jesse Gross wrote:
>
>> It's true, the field ordering is not significant if you understand all
>> of the fields. The concern is that userspace may want to keep track of
>> flows even if it doesn't understand all of the fields (f
On 19 July 2014 06:13, Jesse Gross wrote:
> It's true, the field ordering is not significant if you understand all
> of the fields. The concern is that userspace may want to keep track of
> flows even if it doesn't understand all of the fields (for example, it
> can determine that the kernel prov
On Thu, Jul 17, 2014 at 3:53 PM, Joe Stringer wrote:
> On 18 July 2014 05:19, Ben Pfaff wrote:
>> - If the userspace flow key includes more fields than the
>> > kernel's, for example if userspace decoded an IPv6 header but
>> > the kernel stopped at the Ethernet type, then userspa
On 18 July 2014 10:53, Joe Stringer wrote:
> Basically I'd be much more comfortable with the kernel generating its
>> own flow key the first time it dumps a flow, then caching that. How
>> much overhead does that have versus caching the userspace-provided
>> flow key?
>>
>
> By the same metric,
On 18 July 2014 05:19, Ben Pfaff wrote:
>
> > From datapath/README:
> >
> > - If the kernel's flow key includes more fields than the userspace
> > version of the flow key, for example if the kernel decoded IPv6
> > headers but userspace stopped at the Ethernet type (because it
> >
On Thu, Jul 17, 2014 at 11:04:55AM +1200, Joe Stringer wrote:
> On 17 July 2014 04:48, Ben Pfaff wrote:
>
> > On Wed, Jul 16, 2014 at 02:50:12PM +1200, Joe Stringer wrote:
> > > Converting the flow key and mask back into netlink format during flow
> > > dump is fairly expensive. By caching the us
On 17 July 2014 04:48, Ben Pfaff wrote:
> On Wed, Jul 16, 2014 at 02:50:12PM +1200, Joe Stringer wrote:
> > Converting the flow key and mask back into netlink format during flow
> > dump is fairly expensive. By caching the userspace-provided versions of
> > these during flow setup, and copying th
On Wed, Jul 16, 2014 at 02:50:12PM +1200, Joe Stringer wrote:
> Converting the flow key and mask back into netlink format during flow
> dump is fairly expensive. By caching the userspace-provided versions of
> these during flow setup, and copying the memory directly during flow
> dump, we are able
Converting the flow key and mask back into netlink format during flow
dump is fairly expensive. By caching the userspace-provided versions of
these during flow setup, and copying the memory directly during flow
dump, we are able to support up to 33% more flows in the datapath.
Previous top-5 perf
