Re: Fletcher Checksum Question

From: Craig Partridge (craigp@world.std.com)
Date: 02/05/03 

From: craigp@world.std.com (Craig Partridge)
Date: Wed, 5 Feb 2003 10:21:19 GMT

dank@micrologic.com (Dan Kotlow) writes:

>>Fletcher isnt flawed, but it has limitations, as does any checksum.
>>
>>My implementation may be flawed, however. And my proposed modifications are
>>likely flawed.

>It sucks. I work with a satellite communications service that uses
>it. We get undetected errors frequently. Often the subscriber
>devices seem to get bad control data and go haywire. It is much
>weaker than any decent 16-bit CRC.

Let's be a little careful here.

A nice thesis by Jonathan Stone at Stanford showed that if you don't define
the error model (that is, if you say any change to the packet is equally
likely) then any n-bit checksum is equivalently strong (or weak). You have
to define the error model to make a comparison.

So your statement should be restated to: over a satellite channel, 16-bit
Fletcher performs sharply less well at error detection than a 16-bit CRC.

If the gentleman is using Fletcher in some other environment, he may find
it is better.

Craig