On Tue, Aug 07, 2007 at 05:29:07PM -0400, Dan McDonald wrote: > > I don't think we care about keepalive drop stats, but we should know if > > we do... > > Okay.
I'm guessing we don't :) > > I gather that ESP-in-UDP is fairly costly (e.g., we re-compute the IP > > header checksum for local purposes only). > > For now, we recycle the packet back into existing ESP paths. Something like > PEF will help a lot here (since the "compute header checksum" event comes > before "check ESP-in-UDP"). PEF? > > Probably no big deal since the systems using this are hardly HPC systems > > and probably have bottlenecks other than the CPU, but maybe a comment is > > worthwhile. > > Agreed. Also, it's possible I can recompute the checksum incrementally. Nah. I don't need that and I doubt anyone else does. And I just noticed that the ovbcopy() is just to move the IP header (doh, not the ESP payload) so who cares. _______________________________________________ networking-discuss mailing list [email protected]
