cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
555
Views
0
Helpful
1
Replies

MPLS MTU

jpl861
Level 4
Level 4

Hi Experts,

Just like to ask first your inputs about the MTUs needed on our proposed setup. We currently have a large internal network composed of several metro ethernet links. We have different carriers and we all know that they do not always provide L1 connectivity. They sometimes do Q-in-Q or EoMPLS or other technologies that would hide their internal network and appear as a point-to-point ME circuit to customers.

We are planning to create our own MPLS network for our clients so we don't have to leak their networks inside ours and we are trying to avoid the overhead of GRE/IPSEC since we'll be adding a lot of client networks and the overhead is not reasonable. So we just thought of MPLS-VPN to at least reduce the overhead and we don't have to purchase a lot of network devices.

With that said, what is the safest thing to ask the carriers and what settings should i put in our network devices. I am still confused with the differences of MTU, IP MTU, and MPLS MTU.

We also have one circuit running 802.1q instead of using routed-port on the switch. Is 802.1q supported in LDP?

Our internal network is comprised of 6500 switches with Sup720 and Gigabit linecards and we are planning to use 3900 routers as PEs. We all hooked up our ME circuits across the 6500 switches.

Thanks in advance.

JL

1 Reply 1

jpl861
Level 4
Level 4

Just an update. I tried to do some testing on my GNS3 MPLS network to validate the MTU settings that need to be used.

However, I'm not sure on how Cisco IOS handles ICMP when the packet size is defined but I at least found some interesting details.

I used 3725 routers as CE, PE, and P. All of the LAN links used the default MTU which is 1500.

Test#1:

1 . Ping CE1 loopback to CE2 loopback with 1500bytes and DF-bit set. I got cannot fragment ICMP message on my ping.

2.  Ping CE1 loopback to CE2 loopback with 1492bytes (8 bytes for 2 MPLS labels) and DF-bit set. I received end-to-end connectivity.

Test#2:

1. Increased all FastEthernet/LAN segment to MTU 1508.

2. Ping CE1 loopback to CE2 loopback with 1500bytes and DF-bit set. I got end-to-end connectivity.

3. PIng CE1 loopback to CE2 loopback with 1501bytes and DF-bit set. I got cannot fragment ICMP message on my ping.

Test#3:

1. Retained all FE/LAN segment to MTU 1508

2. Change one of the P to P segment to 802.1q.

3. Same result as Test#2.

Test#4:

1. Retained all FE/LAN sgement MTU to 1508

2. Ping CE1 to PE1's p2p circuit using 1500 bytes and DF-bit enabled. Result is ok.

3. Ping CE1 to PE1's p2p circuit using 1508 bytes and DF-bit enabled. Result is ok.

4. Ping CE1 to PE1's p2p circuit using 1509 bytes and DF-bit enabled. ICMP failed.

So now I'm a bit confused on what settings are really needed. If I configure 1508 bytes as interface MTU and I sent 1500 bytes ICMP, then at the middle of the cloud, two MPLS labels were added with a total of 8 bytes so it would total 1508 bytes. So with the interface MTU configuration, the router ignores the additional 18 bytes header for Ethernet as it will automatically consider it before transmitting data? To complicate things, I tried to change one of the links in the middle of the cloud to 802.1q but left the interface MTU to 1508bytes. I can get end-to-end connectivity with 1500bytes and DF-bit set but not with 1501.

So now, I am assuming that setting the interface MTU would consider the data payload, tcp/udp headers, plus IP header without the layer 2 headers. Then the layer 2 headers are automatically assumed by both ends of the WAN circuit. Not sure if my analysis is correct.