MTU size / effect on TCP / UDP

Hi,

I have a few questions about MTU that I thought somebody on here might be able to answer.

The first is regarding where the actual size of packets sent through the 'internet' is 'decided'. If you can change the MTU on an individual machine (using TCP optimiser for example), but also on the router itself, how is this different? Does changing the MTU on the router force all devices to run at that same MTU? Is it possible for a machine to use an MTU of 500 if the router is still set to default 1500?

Does MTU size determine the size of both TCP and UDP packets? Or is it only TCP?

As you may or may not know, many, many players feel that lowering the MTU can improve network related performance in some fast paced online games. Specifically, when playing Battlefield 3, lowering your MTU can (supposedly) greatly improve hit registration. I tried it myself (lowering the MTU on my machine to 500 using TCP optimiser) and I have to say, it does seem to improve it.

However, being sceptical as I am, I thought I would ask here as to whether it is even posible that simply changing the MTU on my machine is able to change the stream of data being sent to a server running an online game that uses UDP?

Any insight greatly appreciated,
T

it affects both.

try to avoid using 3rd party tools to change these settings as you dont know what else they change.

Google
"fragmentation mtu test"

You just need to ping the address of the server you connect to for gaming. Once you have fragmentation free pings you have found your optimum MTU size. Making it smaller is (in my network engineer profession education and experience) is pointless. Read up on Path MTU too.

If anything you are increasing overheads for more packets being sent by your NIC and routing equipment.

Would love to be proved wrong so I can learn how reducing packet size to a ridiculous size of 3ould help gaming. Also, I would expect that if it was "faster" it would be soo minimal that it wouldnt be noticeable and worth the effort of spanking your hardware.

Internet "standard" is 1500 however your service provider will most likely steal a few bytes for overheads (VLANS,DSCP) hence recommendation to use something like 1472 and the likes.

Did you Frag score leap from 1-100 to 100-1 after changing MTU??

Blue

the simple answer would be to packet sniff it and see what if anything happens to the packet sizes.

Devices can have an MTU which caps what they can send, but if its bigger than the MRU of the thing at the other end, or anything in between, it'll get fragmented and slowed or rejected. PMTUD is intended to find the most constrained part on the path, and control the packet size to fit through that, but it doesn't always work.

You can see the effect of size on ping time :-

phil@minuet:~$ ping www.demon.net -s 1472
PING www.demon.net (89.145.83.36) 1472(1500) bytes of data.
1480 bytes from 89.145.83.36: icmp_seq=1 ttl=57 time=46.9 ms
1480 bytes from 89.145.83.36: icmp_seq=2 ttl=57 time=45.7 ms
1480 bytes from 89.145.83.36: icmp_seq=3 ttl=57 time=45.8 ms
^C1480 bytes from 89.145.83.36: icmp_seq=4 ttl=57 time=45.8 ms

--- www.demon.net ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 15338ms
rtt min/avg/max/mdev = 45.782/46.095/46.908/0.538 ms

phil@minuet:~$ ping www.demon.net -s 472
PING www.demon.net (89.145.83.36) 472(500) bytes of data.
480 bytes from 89.145.83.36: icmp_seq=1 ttl=57 time=37.1 ms
480 bytes from 89.145.83.36: icmp_seq=2 ttl=57 time=36.4 ms
480 bytes from 89.145.83.36: icmp_seq=3 ttl=57 time=35.4 ms
^C480 bytes from 89.145.83.36: icmp_seq=4 ttl=57 time=36.4 ms

--- www.demon.net ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 15228ms
rtt min/avg/max/mdev = 35.426/36.369/37.141/0.627 ms