how much is MTU a factor in speed?

slimothy

Banned
Joined
Jan 14, 2005
Messages
4,808
Reaction score
0
As I understand it the MTU fine tunes the size of packets sent/recieved or something. I know how to calculate the optimal MTU etc etc but how much is it a factor in having good speeds? Does it just squeeze an extra few KB/s in or can it mean the difference between 20KB/s and 120KB/s?
 
I would like to know as well - I have changed mtu's rwin and all other network settings I know and it makes squat difference to my speed...which is still 8 KB/s.
 
The wrong MTU can produce difficulties loading certain sites, but generally it is a tweak that makes your connection more efficient. If you have MTU discovery "on" (which is the default) it will first try each site at your MTU setting, and if incorrect will resend at the "discovered" MTU. Clearly that makes it slightly less efficient than if your MTU setting was correct in the first place. But for downloading a large file, for example, it will have negligible impact.

The RWin setting can also improve efficiency especially on a platform like Iburst which has higher latency. With the right setting it can significantly reduce the amount of data being resent if packet errors occur. Of course, if no packet errors are being encounterd it makes rather less difference.

Neither setting will overcome a poor signal, so getting the best possible signal should be one's first priority.
 
okay folks. here's my definitive (IMHO) definition (Redundant) of the effect of MTU - the maximum transmission unit for a network interface, like ethernet or iBurst radio.

If your connection is configured for a MTU that is higher than your link connection can support, the packet will be *FRAGMENTED* that is, split into two (or more packets if your link is especially bad).

The issue is, your iBurst can only transmit and receive a certain amount of packets a second. This is where a skewed MTU setting can affect speed. If you inefficiently send packets, speed will be affected, and some websites will associate you with a ddos attack, which is quite undesirable.

Example: You're connected to a bittorrent tracker, and peers contact you for data. As you send a packet of data to the peer, whereever he is on the internet, your PC sends a the maximum sized packet of data for your ETHERNET connection, which by default is 1500 bytes including TCP headers.

As the packet reaches your router, iptables or your other NAT software running on the router will say, 'hey man, this packet is 1500 bytes, but the iBurst connection can only send out, by default, packets of 1432 bytes at a time - I'll have to split this into two packets to send this.' Obviously, it takes more time to send two packets than it takes to send one.

The most important problem arises if the MTU on your router is set to the default setting, but this IS STILL TOO HIGH. For example, the default linux MTU for PPPoE (eg Iburst) is 1432, but my radio signal will only support packets > 1350 in size.

So what happens in the logic of NAT at your router is: 'I recieve an outgoing packet of size 1500 bytes. This is too large for the default setting, So I'll split it into two packets - one 1432b in size, the other 68 bytes in size. But hang on, the iBurst router just told me that the 1432bytes packet is still too large for transmission - I'll HAVE TO FURTHER SPLIT THIS PACKET.'

So what happened essentially is that the one packet that you tried to send actually was split into 3 packets where one could have done fine.

Author's note: I'm not sure where the second fragmentation occurs: on the router, or on the UT.
 
DaveBuchanan1337 said:
So what happened essentially is that the one packet that you tried to send actually was split into 3 packets where one could have done fine.
OK, but what significance does it have on my connection speed? Will I even notice a difference in speed?
 
slimothy - I don't have enough information to answer that.

Can you tell me:
How many packets/second can the network support with a low contention ratio

How much data per second per user can be allocated? I know that the maximum throughput in kb/s is 1024 theoretically, but is that 1024kb/s/user leaving the BS to the main switch, or 1024kb/s from the user to the BS?

I don't know enough about the architecture to postitulate.
 
MTU has very little if any influance on speed. The only thing MTU does it control packet size (as pointed out by Dave) and even in Windows it's dynamic now days so it really doesn't matter much.

If you have 8k/s, if those 8k/s sec uses 100 or 1000 packets it's still 8k/s. The only way it might influance speed is if it drops very low (IMHO < 800) and the minor network over head (talking tiny amounts here) starts to add up then maybe it will cause it to slow down since it has to send more data (again it's just very small amounts of data extra).

The biggest reason for fiddeling with MTU is some places still live in the dark ages where they use routers which can't handle fragmented packets and then don't route them correctly which means some sites/programs will not work correctly.
 
exactly so then why are the support guys always getting everyone to fine tune thier MTU like its some magical setting that will fix iBurst
 
slimothy said:
exactly so then why are the support guys always getting everyone to fine tune thier MTU like its some magical setting that will fix iBurst
I think we would all like to know the answer to that question - hint hint ;).
 
no clue. might be the old support trick of, have the user change a million settings, asking them to try it for a few days between each change, will we stall for someone to fix the real problem.
 
Top
Sign up to the MyBroadband newsletter
X