TNSR and NAT
-
I'm really excited to start seeing TNSR on bare metal and have been doing as much reading as I can find on TNSR and VPP(fd.io). Even thought of trying to load up an instance of VPP on my own to do a little testing but I doubt I have the sysadmin skills to pull it off without wasting days-weeks on it.
I've come across some of the regression testing results for VPP, focussing on the NAT performance results, as I understand it seems that the tests are based on 1000 sessions established through the NAT. I was wondering if any testing has been done on something similar to an XG-1537 or 1541 with TNSR to check for session(state x2) overload to see how many millions of active sessions can be maintained? Is there a point where more memory becomes irrelevant? Can you double your state capacity by SNAT to NAT44(LSN/CGN)?
Lots of questions :) excited to see what the future holds!
-
Hi,
We're excited for TNSR on bare metal too!l As bare metal version still in development, we won't be publishing any test results just yet. Do keep an eye on our social media as we will reveal more information once it's ready. Thanks!
-
We haven’t done any performance testing of NAT yet.
That said, some of the CSIT performance tests don’t just test 1,000 sessions. Some of them test 4000 users (local source IPs) with 15 sessions each.
https://docs.fd.io/csit/rls1804/report/detailed_test_results/vpp_performance_results/vpp_performance_results.html#ge2p1x520-ethip4udp-ip4scale4000-udpsrcscale15-nat44-ndrpdrdisc
That particular test yields 6,588,541 PPS (64-byte frames) with (and this is important) zero dropped frames.
-
Damn, thats pretty good! 3.73Gbps with smallest frames possible and 80Gbps at full sized frames :)
Looking forward to release and seeing where the limits are!
-
@80211wiguy it's 4.505362386 Gbps (64-byte frames).
-
@jwt oops
