1

Let's say, that one has a following network topology:

www-server[eth0] <-> [ge-0/0/0]MX-router[ge-0/0/1] <-> Internet

What are the advantages of setting the Maximum Transfer Unit and Maximum Receive Unit at the web server-facing router Ethernet port? I guess if the MRU is set to 1500 bytes, then even if the server Ethernet port is set to >1500 bytes, then frames larger than 1500 bytes get dropped as quickly as possible. Otherwise they would probably get dropped somewhere at the client infrastructure when they hit the 1500 byte limitation because in service-provider core the MTU is larger than 1500 bytes. And I guess the 1500 byte MTU serves the same purpose? For example, if somehow >1500 byte packet addressed to web-server reaches the router, then for IPv4 it fragments it and for IPv6 it sends a ICMPv6 "packet too big" message back to client.

2
  • 1
    I'm not sure I understand your question. You are not going to get a packet larger than 1500 octets from the public Internet. The ethernet MTU is 1500 octets. There are some devices that can use jumbo frames, but it is non-standard, must be explicitly configured, and different vendors apply it differently. Using jumbo frames, the MTU would need to be the the same on the entire path from source to destination, and you will not be able to send packets larger than 1500 octets on the public Internet.
    – Ron Maupin
    Nov 23 '18 at 16:45
  • @RonMaupin Ok. I simply imagined a (unrealistic) situation, where there is a service provider which configured its infra in a way that all the ports, including the customer-facing ones, support jumbo-frames and thus if their clients configure their end-devices to support >1500 byte MTU, then in theory, my router would also route those packets if port facing the web-server has jumbo-frames enabled.
    – Martin
    Nov 23 '18 at 18:19
3

For IP over Ethernet, the MTU is generally 1500 bytes.

However, if your Internet uplink only supports a lower MTU (e.g. 1492 bytes with PPPoE or L2TP) your router is forced to fragment each full-size packet that the server sends towards the Internet. To remove that additional processing and bandwidth overhead you can tell your server to reduce its interface MTU to (in that case) 1492.

The MRU can sometimes be set separately from the MTU to allow reception of standard 1500 byte packets while sending slightly smaller packets.

On the other hand, there may be scenarios where your path between server and client is able to support a larger MTU and you want to lower the IP and Ethernet processing overhead. Then you can consider deviating from the Ethernet standard and use larger, jumbo frames.

Note that jumbos are generally non-standard, only work if all devices on the segment support that frame size, and they can cause significant overhead when fragmented across a router. Jumbos only make sense in a very controlled environment and usually require extensive testing.

Also note that although "MTU" is an IETF term for the layer-3 IP packet size, some vendors use "MTU" to instead refer to the maximum layer-2 frame size. This can create severe problems since the layer-2 header needs to fit in the frame as well. Check the manuals thoroughly or contact support as to which exactly is meant. Also, a default "MTU" value of 1518 or 1522 bytes refers to the maximum frame size and not the L3 MTU. When in doubt, set the maximum frame size to a slightly higher value on the switches - they don't mind if no sender is actually using that size.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.