IPv4-only networks via IPv6-only network
Ichiro Okajima
okajima at mlab.yrp.nttdocomo.co.jp
Thu Dec 4 12:09:46 CET 2003
Hello Guus,
> > But, I run into another problem. The round trip time of tunneled IPv4
> > packets between two tincnized Redhat9 PCs fluctuates on the order of
> > 0.532 to 10.554 ms, although that of IPv6 packets between them are
> > very stable. Because I am planning to use a VoIP application between
> > the two IPv4 network, stable RTTs of IPv4 packets are necessary.
> >
> > Do you have any idea to improve the RTT?
>
>Since tinc is a normal user-space program, it is subjected to the task
>scheduler, which on i386 architectures normally switches tasks every 10
>ms. So if there's another non-idle program running on those PCs, tinc
>might not respond as quickly as when it's the only process running. You
>can change this behaviour by giving it a higher priority with nice, or
>you can let tinc run in realtime (use the chrt command from the
>schedutils package for example).
According to your advice, I set higher priority to tincd with nice
command. It seems that it makes RTTs more stable than before.
> > --- 192.168.20.2 ping statistics ---
> > 100 packets transmitted, 100 received, 0% loss, time 99118ms
> > rtt min/avg/max/mdev = 0.532/3.112/10.554/3.093 ms
>
>On the other hand, it's only a 3 ms standard deviation, and a worst case
>RTT of 10 ms. Doesn't seem that bad to me, even for VoIP!
Unfortunately, VoIP application I have to use is very sensitive
to the fluctuation of RTTs, probably because the play out buffer
mechanism is not intelligent.
Thanks,
Ichiro Okajima
Tinc: Discussion list about the tinc VPN daemon
Archive: http://mail.nl.linux.org/lists/
Tinc site: http://tinc.nl.linux.org/
More information about the Tinc
mailing list