[ovs-dev] Userspace Netlink MMAP status

Thomas Graf tgraf at redhat.com
Tue May 6 10:14:46 UTC 2014


On 04/29/2014 09:47 PM, Ethan Jackson wrote:
>> The common use case is a flow expiring during the lifetime of a TCP
>> connection. It will result in multiple data packets being sent upwards.
>> It's much less likely in the megaflows era though.
>
> I think this is pretty unlikely except in some extreme circumstances.
> For a flow to be removed from the datapath, it needs to be idle
> (unused) for about 1.5s.  One could construct pathological cases in
> which this would be the case for bursty high bandwidth flows, but I
> highly doubt it's typical.

Back from PTO ;-)

I agree that it's non typical but it's certainly exploitable, especially
if slow start after idle is disabled. The fact that slow start after
idle was requested to be configurable over and over again indicates
that there are uses cases that let connections idle followed by bursts.



More information about the dev mailing list