09-17-2015 06:11 PM - edited 03-05-2019 02:20 AM
Hello Everyone,
I'm trying to better my understanding of MTU, and I can't seem to find the answer through research.
A client of mine has encountered a performance problem, which may be related to an MTU mismatch. However, I'm having a hard time confirming.
I have a router that has its outside (WAN) interface MTU hard set to 1500.
The router interface that's directly connected (the ISP) has its MTU hard set at 2000.
From what I've gathered, we now have an undesired MTU mismatch.
However, I'm trying to see if it's actually causing the performance problem. We occasionally receive jumbo frames on our interface, but I'd like to know what's actually happening with the MTU. My understanding, is that one of two things happens to counter-act dropped packets. Either the sending host (behind my router) will use RFC 1191 (MTU path discovery) and realize it needs send packets with an MTU of 1500 (or less) through this link. Or the router is going to fragment the packets.
My questions:
Is MTU calculated twice here. Once by the sending host, and once again by the receiving host?
If I'm understanding, both will monitor this session with their own MTU path discovery value.
Would you suspect this configuration to be a problem?
Do you think this is a common ISP configuration?
Most MTU issues I've had are UDP and IPSEC related. I feel like this shouldn't be a problem, but I'm second guessing myself.
Thank you for any additional insight into this.
It's much appreciated.
09-17-2015 07:20 PM
Hi Nathan,
From your description, Am I assuming right that you have a host connected (on LAN interface) of the router. The WAN interface of this router is with 1500 as MTU and connected to ISP router (and this interface is set to 2000 as MTU)?.
If so...
My questions:
Is MTU calculated twice here. Once by the sending host, and once again by the receiving host?
<Nagendra> PMTUD is unidirectional and relies on ICMP "Packet Too Big" (fragmentation required) messages. So it is not negotiated between sender and receiver. Sending host and receiving host will discover the path MTU. This is required because, the path taken by packets from Router-A to Router-B is not required to be the same path for packets from Router-B to Router-A (asymetric or ECMP networks for example).
If I'm understanding, both will monitor this session with their own MTU path discovery value.
<Nagendra> Right.
Would you suspect this configuration to be a problem?
<Nagendra> Normally yes. The best practise is to have same MTU on both the router. In your topology, if ISP router is receiving a packet of size 1900, it will forward to your router. Since in your router, MTU is set to 1500 on incoming interface, it might drop the packet. To be specific, this will not generate ICMP error message (as the drop happens as L1/L2 layer) and PMTUD may not work. So it is better to have same MTU value.
Do you think this is a common ISP configuration?
<Naendra> Not that I am aware of. It is a common practice to have same MTU value.
-Nagendra
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide