MPLS & increasing MTU size

[SJerseyRob]

I'm fairly new to Cisco and I've been reading about the need to increase the MTU size on (Gigabit)Ethernet interfaces in an MPLS network. I believe I'm pretty clear on the need to increase MTU size on (Gigabit)Ethernet interfaces on the CE router. (Just looking at the customer side only for this discussion.) But I'm not clear from my reading about the need to do that on the LAN-connecting (Gigabit)Ethernet interface on the CE router connecting the CE router to the (Gigabit)Ethernet ports on the distribution switch? MPLS PE router <--> GigE 0/1 [Customer CE router] GigE 0/2 <--> GigE 1/1 [Catalyst 3760] GigtE 1/2 <--> LAN I'm thinking (for example) that GigE 0/1 on the CE router should be configured like this: ! CE(config-if)# mtu 1524 CE(config-if)# mpls mtu 1524 CE(config-if)# ip mtu 1500 ! (1) Should that config be repeated on GigE 0/2 on the CE router? (2) Also on GigE 1/1 on the 3760? (3) Also on GigE 1/2 on the 3760? Thanks for putting up with what is probably a very basic question! Rob

[Georg Pauwen]

Hello,

as I understand it, non MPLS enabled interfaces do not need to have their MTU size changed; the default MTU size of 1500 is still applicable to traffic coming from the end hosts.

[Joseph W. Doherty]

Ditto.

MPLS, like VLAN tagging, automatically extends the size of the frame so, as far as I know, you don't need to configure an increase in the size of the MTU.

[SJerseyRob]

Joseph - thanks for responding!

[SJerseyRob]

Georg

Thanks for responding. I think you've answered my ultimate questions - should the MTU size be increased on GigE 0/2 on the CE router and on GigE 1/1 and GigE 1/2 on the distribution switch.

Just to be sure, (A) do you agree that MTU needs to be increased on GigE 0/1 on the CE router since that's a MPLS enabled interface? (B) You mention traffic coming into GigE 0/2 on the CE router *from* the end hosts. Is your answer the same for traffic exiting GigE 0/2 on the CE router *to* the end hosts?

Just so you know, here's some of the sources that got me thinking about MPLS and MTU size:

1. http://wiki.kemot-net.com/mpls-mtu

Introduction

MTU in MPLS can be an issue since each MPLS label imposed on a frame decreases the transported protocol's maximum size.

MPLS MTU

• MTU of an interface defines the maximum Layer 3 packet size that can be transmitted.
• Each labeled packet adds 4 bytes for each MPLS header. MPLS VPN uses two labels, 2*4 = 8 + 1500 = 1508 bytes.
• The default MPLS MTU is 1500.
• Knowing the max number of labels, you can set the MPLS MTU with the command mpls mtu (value). For example 2 labels should have a MTU of 1508 to support 1500 IP packets.
• For packets that exceed the MTU, MPLS strips the labels, fragments packets and puts back labels for each fragment . . ..

2. http://www.itino.net/mtu-size-in-a-mpls-network/

MTU size in a MPLS network

MTU size never crossed my mind until MPLS designed network came to the picture. After adding a switch to the network, OSPF got up, but BGP was going up and down and up and down… That led to the MTU mismatch because bigger frames are needed in a MPLS network.

The default MTU frame size is 1500 bytes, but in a MPLS design where you use labels, with every label you need a little more room:) For the one label design you need to increase the MTU size to 1504 bytes. MPLS VPN, which is the most common design, uses two labels and thus you have to increase the MTU size to 1508 bytes or by eight bytes. Each label increased the size of the original IP packet to a value bigger than the original MTU size can handle. In a default MTU size, configuration packets that are bigger get fragmented and when such packets are sent with the Do-Not-Fragment (DF) bit set, packets get dropped by the router.

Thanks again for your help!

Rob

[Georg Pauwen]

Rob,

indeed it only need to be increased on the interface facing the MPLS router, GigabitEthernet0/1 in your case. On Ethernet interfaces, I think you need to configure ' mpls mtu override bytes'...