Hi,

I have some border routers running x86 version of routerOS. A few weeks ago i posted about the current mpls ldp implementation that only add on label to ecmp routes so its broken ecmp traffic. Then i worked arround and with Te ( RSVP) i was able to run bgp based vpls tunnels. With the default configuration the tunnel get R ( running) but only small packets worked, after running a lot of testing with packet sniffers, i figured out that the problem is that the X86 router cant use anything bigger than 1500.


the testing setup was a x86-----rb450G----rb450G and then running a BW test and packet sniffer.

the VPLS traffic between x86 an the last rb450g doesnt work with mpls mtu bigger than 1500 ( wireshark 1514 bytes on wire) so we get a lot of fragmentation that hurts the RB CPU.

later then i tested a vpls between the two rb450 g and its worked like a charm with a mpls mtu of 1522 ( wireshark 1536).


Our X86 server run intel pci-x server nic that in documentation can run jumbo frames without problems.

anyone experienced a problem like that?

On X86, if ethernet card supports jumbo frames, you can set MTU larger than 1500

On X86, if ethernet card supports jumbo frames, you can set MTU larger than 1500

The problem is about MPLS MTU or (l2MTU reading) if i set anything bigger on MPLS MTU , all frames generated from a vpls from a RB450G for example to the X86 Machine will be discarded. After some time testing it, i discovered that MPLS interface MTU on x86 doesnt work, even if a set it to 1600 ( the nic is an Intel Server that supports JUMBO Frames). Because traffic generated via vpls from X86 works, but traffic generated by rb450g with mpls mtu of 1522 doesnt work . The only packets that goes trhu is small packets that doesnt. and the only way to the x86 to not discard the vpls packets generated by rb450g is setting mpls mtu on 450g to 1500.

On x86 it is not possible to change L2MTU, if NIC supports jumbo frames them L2MTU can be changed by setting MTU of an interface.

So if you want to forward 1522 byte packets, you have to:
* set MTU of an ethernet interface to 1522
* set MPLS MTU to the same value as in other MPLS enabled routers which should be less or equal to L2MTU.

Thanks MRZ, gonna try that workaroud in LAB. Any ETA about X86 work as Routerboards that we only need to change MPLS MTU to get packets bigger goes trhu? So we dont need to mess with Ethernet Interface MTU?

Because as you know if i change Interface MTU Ospf will broke because its checks the mtu. As we have a quite large network with about 50 Points of presences running routerboards and Our core routers are X86 , if we change the MTU to 1522 on x86 , the ospf will go down as its a production network we cant do that without causing problems with our enterprise clients.

It is not possible to add L2MTU for every NIC in the world. We will consider adding a feature similar to Ciscos "ospf mtu-ignore"

That will help!!!
Thanks!

It is not possible to add L2MTU for every NIC in the world. We will consider adding a feature similar to Ciscos "ospf mtu-ignore"

If you use ethernet cards which are built around the same ethernet chips you can find in Routerboards (eg some VIA Gigabit chips),
you can set L2MTU in x86 based Routers! The only trick is to find them