Issues with max MTU on pings to Amazon AWS

[TonySuffolk]

This might not be a BT network issue but I need to ask here.

I play Elite Dangerous which at times is very susceptible to unexpected packet fragmentation, and has it's own code for splitting data across multiple UDP packets.

The game itself is hosted on the Amazon AWS and one of the servers in London (the closests to me) at IP address 3.8.0.0

I use the command

ping -f -l <length> 3.8.0.0

to find the max mtu that the route will accept  - initially  I found that a length of 1000 bytes was sufficient.

Within about 30 minutes of playing the game crashed, and when i tried the ping again the Max MTU on the route was down to 900, and then dropped again to 850.

I assume that the changing MTU and subsequent fragmentation is down to different routes being traversed from my network to the Amazon AWS, but is that likely to be a BT network issue or a more general networking problem.

[ptrduffy]

I can't imagine that the game developers expect you to set the MTU to make your game work.

Regardless of that, using the '-l' option on a ping request tells you nothing about what happens to UDP packets with a certain length.  Ping uses a protocol called ICMP (Internet Control Message Protocol), specifically an ICMP Echo Request, which is entirely separate from UDP.

So whatever is happening on your game, the ping is not useful for diagnosing the issue.

[TonySuffolk]

All I can tell you is that I have been told by the support team for the developers have told me to use ping -f -l .... to find the optimal MTU.

The developers have said that it is important to ensure that the fragmentation isn't an issue and one of the trouble shooting tips is to set the Max MTU - the expectation they have I am sure is this is going to be a one time only setting.

The point is that in reality the max MTU keeps changing minute to minute - even today within minutes I had packets of length 680 being fragmented, and then packets of size 700 being accepted.

Regardless of whether this is relevant to the game specifically (obviously the developers consider it to be useful) is it normal for a network route to be this unstable  in terms of MTU sizing and fragmentation ?

[licquorice]

@ptrduffy  Ping <url> -f -l xxxx will reveal if the packet needs to be fragmented or not.

However, I agree that setting the MTU should definitely not be something that is expected of a user

[ptrduffy]

Sorry, my previous reply was wrong - I blame lack of caffeine, too early in the morning.  The MTU is an IP layer property, and both ICMP and UDP will have the same limits.

Having said that, there is no guarantee that packets between you and any server will take the same route, so they could always be subject to changing MTUs (although the numbers you are reporting so seem low).

I'm not very familiar with ping on Windows machines - I can see that the '-f' option is "do not fragment" (on my Mac it's a "flood" option), but how are you determining that the packets are being dropped because of this (as opposed to just being dropped due to a host being too busy, or being configured that way)?

[TonySuffolk]

thanks for the reply.

it isnt that packets are being dropped but that for some reason the game  code struggles with fragmented traffic.

Why i don't know, but the developers themselves have said that fragmentation causes problems in game.

It is curious since fragmentation and re-assembly is a solved problem.

[licquorice]

Sledgehammer to crack a nut, but an MTU of 576 is guaranteed to be reassembled if fragmented.