Tinc performance on a Dir-300

[Rob Townley]

On Mon, Sep 20, 2010 at 11:27 AM, Guus Sliepen <guus at tinc-vpn.org> wrote:
> On Mon, Sep 20, 2010 at 04:37:03PM +0200, Clemens John wrote:
>
>> we are using Tinc in our Freifunk Network in Oldenburg for internode
>> connections over the internet. So Tinc is running on OpenWrt 10.03 on Dlink
>> Dir-300 Routers.
>> We all have enough internet bandwith (1,6 MB/sec and more) but we only get a
>> maximum speed of ~350KB/sec between two tinc nodes because then tinc uses 99%
>> of the cpu.
>>
>> Is it possible to get more Speed with tinc on this machines? I think we have
>> compression and encryption already turned off so what is using the cpu?
> [...]
>> If there is no way to get more speed, do you know another VPN-Solution which
>> is better concerning speed? We dont need security because the network is
>> completely open, but we need speed.
>
> Are you sure you have 1.6 MB/sec both upstream and downstream?
>
> You can try other VPN software, such as OpenVPN or perhaps VDE, and see if
> those are faster.  Maybe you can also try profiling tinc (either with
> valgrind's callgrind tool, or compile tinc with -pg in both CFLAGS and LDFLAGS
> so you can use gprof) to see where the bottleneck is.
>
> However, the Dir-300 has a MIPS processor running at 182 MHz with only 16 kB
> instruction and 16 kB data cache. That means it is not very fast, especially
> not if a packet has to go all the way to userspace to be processed by tinc and
> back.
>
> --
> Met vriendelijke groet / with kind regards,
>     Guus Sliepen <guus at tinc-vpn.org>
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.10 (GNU/Linux)
>
> iEYEARECAAYFAkyXi2AACgkQAxLow12M2nsYMgCeKWMubZfL1xpJrpaNEe0DKihi
> RLcAoIkxbuk0JU2SADcZ5uRNdUyvpuEI
> =3Owl
> -----END PGP SIGNATURE-----
>
> _______________________________________________
> tinc mailing list
> tinc at tinc-vpn.org
> http://www.tinc-vpn.org/cgi-bin/mailman/listinfo/tinc
>
>


Have you tried nice?  On a full PC platform running Fedora 13 our
gateway tinc node does much much better with nice.  Even though there
were plenty of unused CPU cycles,  changing /etc/init.d/tincd by
prepending 'nice -n 20' to tincd --net=YourNetName
made performance much more consistent.  Otherwise, ping times would
range from 2milliseconds to 4000milliseconds.