[nsp] AToM

Lars Erik Gullerud lerik at nolink.net
Tue Dec 23 06:24:59 EST 2003


On Tue, 2003-12-23 at 11:42, Sergey V. Artjushkin wrote:
> > > But I had a problem with big packets between two PC that was connected
> > > throught trunk port to my cisco devices.
> > 
> > Keep in mind that MPLS normally requires at least 8 bytes extra, and
> > EoMPLS has further encapsulation overhead.
> 
> I know about additional bytes because of VLAN and MPLS tags.
> Let's count the maximum size of the packet.
> One PC sending 1500 bytes to other PC.

Actually 1518. 14 Ethernet header, 1500 payload, 4 FCS.

> When the packet arrives at the switch, switch adds 4 bytes VLAN tag.

Right. 1522 now.

> When packet arrives at the PE router (6509/7401) it adds 8 bytes (2x4 bytes)
> MPLS label tags. So total packet size is 1500 + 4 + 8 = 1512 bytes.
> Right? 

No. You now have 1522 + 12 bytes MPLS - 4 bytes FCS. 1530 now.

> All input/output ports of the switches was configured with MTU=1540 bytes.
> But big packets didn't reach remote PE. 

If 1540 is in fact the physical MTU, you are short. The 1530 byte packet
is in itself a payload that is encapsulated in a local ethernet frame,
meaning another header+FCS will be added, i.e. another 18 bytes. The
actual ethernet frame you are transmitting over the wire is now 1548
bytes.

If however 1540 is actually IOS-style MTU (e.g. you are referring to the
actual Cisco switches here, no other switches involved), which only
counts the payload, then 1540 should have been sufficient.

/leg




More information about the cisco-nsp mailing list