The basis of this is that there is a standard MTU for each type of
layer 1/layer 2 network. For ethernet the default is 1500, for PPP
it is 576. Now the problem is a bug in Win95 DUN that sets the MTU to
1500. Each time a transmission is made it tries sending it with an MTU
of 1500 and fails, and then must downshift to 576. If you adjust the
MTU to 576 to begin with this downshift never happens.
I did do this on my own machine and several others and have seen a
tremendous speed increase. It is possible (as I have done) to use
a modem log file and the Windows system monitor utility to actually
measure the improvement.
The easiest way to adjust the MTU on Win95 is a utility that I picked
up called TweakDUN. It is available at
http://www.gulftel.com/~pattersc/tweakdun/
This program saves the trouble (and possible risk) of hand editing
the registry..
Tim Hodges
7thodges@3rivers.net
At 08:38 AM 8/6/97 -0700, you wrote:
>At 11:52 AM 8/3/97 -0600, Jawaid Bazyar wrote:
>>On Sun, 3 Aug 1997, Jeff Halper wrote:
>>> I was told i could increase the user's speed on the net by changing the
>>> MTU to 576. Any one have any feedback on this??
>>
>>Only under the following conditions would decreasing the MTU result in an
>>improvement in performance:
>>1) User's phone line is noisy
>>2) User does not have an error correcting modem
>>
>>Otherwise, you've just tripled the IP header overhead, *decreasing*
>>performance.
>
>MTU is the Maximum Transfer Unit for a ppp frame. If a ppp frame is
>damaged in transit due to noise, low flying UFOs whatever, then that packet
>will have to be retransmitted. If you have a noisy line, ppp frames will
>have to be retransmitted more often so reducing the MTU will mean that
>smaller frames will have to be retransmited. On a noisy line this would
>tend to increase performace.
>
>The trade off for a lower MTU is increased overhead of packaging, sending,
>receiving, packaging and processing of an increased number of frames.
>
>If line noise was measurable according to noise frequency and duration (I
>doubt it can accurately be done) then you could come up with an equation
>that would optimize performance by balancing MTU size vs. additional
overhead.
>---
>jstorms@livingston.com
>Diplomacy: The art of saying good doggie
>while seaching for a big rock.
>
>