GRE/gif/netgraph tunnel speed on 10Gbit channel

[Vitalij Satanivskij]

Most likely threading only part of problem with iperf3. 

As I say before - I'm check iperf3 in paralel by running multiple instanse on diferent's ports 
Speed is always same. 

But maybe 2-3 instance to few for noticible result's.

For me most confusion was speed aroud 10Gbit on pure link showed by both iperf and iperf3 ...  

Any way. With enabled pf (set skip on gre0 in config) in kernel I get around 3gbit's and around 4.4gbit's without. 

Also I'm tested diferent mtu on interface (eg 9000 and so on) no speed improvement at all

Now question is - is it highgest speed of tunnel that we can expect from GRE or there is some tuning capabilitis for speed up it?



Muenz, Michael wrote:
MM> Am 29.05.2018 um 16:02 schrieb Rodney W. Grimes:
MM> >> Am 29.05.2018 um 13:33 schrieb Vitalij Satanivskij:
MM> >>
MM> >> MM> Can you try MTU 1400 on GRE0 and MSS of 1300 just for testing?
MM> >> MM> With this values I get 4600Mbit sending and 1400Mbit receiving with
MM> >> MM> iperf and 10 parallel streams.
MM> >> MM>
MM> >>
MM> >> UUUPSS looks like i have problem with metodology
MM> >>
MM> >> I'm use iperf3 and even with -P10 or run's in parallel (on diferents ports) it show summary of +-2gbits max
MM> > Iperf3 is known not to do Parallel well:
MM> > -P, --parallel n
MM> >                number of parallel client streams to run. Note that iperf3 is
MM> >                single threaded, so if you are CPU bound, this will not yield
MM> >                higher throughput.
MM> >
MM> > I suggest using iperf for -P tests that cause the single thread of
MM> > iperf3 to become CPU bound.
MM> 
MM> Oh, this was new to me. With iperf I got consistent 4,6Gbit in both 
MM> directions.
MM> _______________________________________________
MM> freebsd-net at freebsd.org mailing list
MM> https://lists.freebsd.org/mailman/listinfo/freebsd-net
MM> To unsubscribe, send any mail to "freebsd-net-unsubscribe at freebsd.org"