MTU network setting...?

esuhl

A while back I was setting up a router, and wanted to check what value to use for the MTU (maximum transmission unit).

Previous ISPs invariably told me to use either 1492 or 1500. But I read that, by trial-and-error, you can find the largest unfragmented packet you can ping, and add 28 bytes of overhead to find the MTU:

Linux: ping -c 4 -M do -s 1420 8.8.8.8
Windows: ping 8.8.8.8 -f -l 1420

The largest packet I could ping was 1420 bytes, making my MTU 1448. I checked on a friend's PC/ISP and also got 1448.

Q1: Is it normal to have an MTU of 1448? It seems unusual. (We both have ADSL2+ connections.)

Also, from decades past, I remember a program called Dr.TCP that let you change the MTU (etc.) settings on the client PC. ISPs recommended it.

Q2: Is it necessary to set the MTU on client devices? Or was that just for old-style dial-up connections before people had routers and LANs?

Q3: What's the best way to minimise packet fragmentation? Is it inevitable? Is it not really an issue?

Any tips gratefully received!

onomatopoeia99

FTTC uses PPPoE which has a maximum MTU of 1492 because PPPoE imposes its own 8 byte header taking it up to the 1500 byte frame size for Ethernet. There are somewhat arcane ways to get 1508 byte frames to allow a 1500 byte MTU on FTTC on BT 21CN backhaul.


The backhaul can also impose a limit - TalkTalk wholesale backhaul for example, which I use, imposes a limit of 1492 (workaround available with my ISP to 'fix' it to 1508 byte frames giving 1500 MTU or 1492 with PPPoE, but I doubt customers of talktalk retail get that option).


I don't worry about it. It just means the transmission overhead is a little higher. I work with a 1492 MTU over TT backhaul and don't worry about eeking out the last fraction of a percentage of download throughput.



Your MTU seems a bit low but unless it's causing you a problem I wouldn't worry about it. Remember that a ping to google's nameservers exposes the MTU of every router on the way there, starting from the one in your house. Have you tried finding the MTU to your router?

esuhl

onomatopoeia99 wrote: »

It just means the transmission overhead is a little higher.

Oh... I assumed the overhead was a fixed 28 bytes, and the payload would be smaller...?

onomatopoeia99 wrote: »

I work with a 1492 MTU over TT backhaul and don't worry about eeking out the last fraction of a percentage of download throughput.

Your MTU seems a bit low but unless it's causing you a problem I wouldn't worry about it.

I'm not worried it... I'd just like to know why. (For no reason other than I'm curious and like to know how stuff works.)

1492 seems normal; 1448 a bit odd. :-/

onomatopoeia99 wrote: »

Remember that a ping to google's nameservers exposes the MTU of every router on the way there, starting from the one in your house. Have you tried finding the MTU to your router?

The MTU to my router is whatever I set on the client device -- 1500 by default. It seems the router's MTU setting only relates to the WAN connection, and the LAN MTU is determined by the clients.

But since I don't use the LAN (except for Internet access), it "makes sense" to set the MTU on the clients to prevent unnecessary fragmentation, right?

Or... is there some clever detection of the MTU? I've heard mention of "Path MTU Discovery (PMTUD)" and "MSS clamping" and "Blackhole detection", but I don't really understand the practicalities.

onomatopoeia99

esuhl wrote: »

Oh... I assumed the overhead was a fixed 28 bytes, and the payload would be smaller...?

I meant the overhead as a percentage of the total data. It's the same number of bytes per frame, but if the frame is smaller, the overhead makes up a larger percentage of the frame.


And I agree 1448 seems unusual, but unless you want to test the highest MTU you can achieve on each hop between your computer and your destination you won't know where the restriction is imposed, and unless it happens at or before your router, you won't be able to do much about it.


As an example of stuff that can affect it, we had some of those abominations that are "homeplugs" at our old office to fix a gap where we couldn't run ethernet cable and they played havoc with the MTU.



They played havoc with plenty of other stuff as well, to the extent that I came in one weekend and spent it drilling holes in walls and feeding cat5e thorough to eliminate the bridge they were supposed to create, to the vast gratitude of the people at that end of the building :rotfl:.

esuhl

onomatopoeia99 wrote: »

And I agree 1448 seems unusual, but unless you want to test the highest MTU you can achieve on each hop between your computer and your destination you won't know where the restriction is imposed, and unless it happens at or before your router, you won't be able to do much about it.

Ahhh... Thanks.

I had a play around and yes -- the MTU seems to depend on the IP/domain I ping. 1448, 1458 and 1492 seemed to be common values.

Pinging my ISP's DNS (or opendns.org) is possible with an MTU of 1492. So I can be pretty sure that's the "correct" value for the router's WAN setting.

---

I presume there must be some kind of clever technology, whereby networks can discover each other's MTUs and not pointlessly fragment packets...?