[ previous ] [ next ] [ threads ]
 
 From:  Marko Vukovic <marko at aquamanta dot co dot za>
 To:  m0n0wall <m0n0wall at lists dot m0n0 dot ch>
 Subject:  Re: [m0n0wall] Help how change MTU size to 2048 value on WAN inte rface
 Date:  Thu, 09 Feb 2006 21:41:52 +0200
Jeff Buehler wrote:
> Higher MTu's run the risk of fragmenting packets, which slows down 
> performance.  So, you want the highest MTU possible for any given 
> connection without fragmenting ("ping -f -l destination" will tell you 
> if a packet has fragmented).  Standard cable/dsl/t1 connections have 
> MTU's of 1500.  The ping to test this is actually "ping -f -l 1472" 
> which equals an MTU of 1500 because of 28 extra bytes in the header (or 
> something like that) so 1472 is actually an mtu of 1500.

The 1472 is MSS which is maximum segment size. I thought TCP/IP headers 
were 40 bytes.

> Making things more complicated, OS's also set MTU for the adapter 
> (ifconfig for *BSD, for Windows you will need to research registry 
> changes or get a tool to modify the MTU).

Don't forget to reboot your 'doze box after changing MTU in registry ;)

> So, if you had a LAN composed of gigabit devices that support jumbo 
> frames, you should get significantly improved performance by setting 
> your MTU to 9000 for all of the devices in the OS.  Then, for your WAN 

I heard somewhere that 9000 was about the correct size to achieve 1Gbps 
throughput on a 32 bit machine with a Gigabit adapter. My guess is that 
most hardware won't support MTU much bigger than that for a while. I 
believe another popular large MTU size is 4470, used by ATM and SDH.

> connection, you would almost certainly want and MTU of 1500 because that 
> is likely the limit of the DSL/Cable/T1 line that you connect to.

Yep, I'm using iBurst wireless here (ZA) and they recommend 1352. Our 
ADSL here typically uses 1492.

> I hope that helps.

Definitely! So it would seem logically that the less likely it is for 
fragmentation to occur, the larger ones packets can be. I guess its just 
a matter of finding an optimal value.

Cheers!
--
Marko