Re: [Xen-users] MTU limits?

[Ernst Bachmann <e.bachmann@xxxxxxxx>]

On Friday 02 September 2005 17:58, Patrick O'Rourke wrote:
> The network device on our hardware uses a rather large MTU (~16k), but I
> noticed that when setting up bridging, the MTU for the bridge and veth
> device is set to 1500.  Trying to set the MTU on the bridge, or eth
> device to match that of the peth is unsuccessful:
>
> [root@por-hv-proto3-p4 ~]# ip link set eth0 mtu 16896
> SIOCSIFMTU: Invalid argument
> [root@por-hv-proto3-p4 ~]#  ip link set xen-br0 mtu 16896
> SIOCSIFMTU: Invalid argument
>
> We're using xen-unstable, but a very similar issue was discussed
> regarding 2.0.6 on the list a while back, see:
>
> http://lists.xensource.com/archives/html/xen-users/2005-07/msg00187.html
>
> IIRC, the poster was suggesting that the 1500 MTU limit is in the bridge
> or the vif code and I was wondering if this is the case.
>
Afaik the vif code of 2.0.x uses a 4k max ringbuffer for transfering packets, 
and does not handle overflows of that buffer, meaning that a MTU of >4k will 
truncate or drop all larger packets anyways.

But the bridge code seems to limit the mtu to 1500, even with no real 
interfaces attached:

# brctl addbr test
# ip link set test mtu 1600
SIOCSIFMTU: Invalid argument

So, with bridging you're stuck at 1500, with routing it might be possible to 
raise the MTU, but that will most likely reduce overall network performance, 
as the number of truncated packets with invalid checksum will raise...

See one of the recent threads about network hangs, ping packet loss, ssh 
session hanging etc for details...


/Ernst

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users