mycroft@mit.edu (Charles M. Hannum) writes:

> Unfortunately, both of these implementations are substantially more
> complicated and less efficient than they could be, because they try to
> hack around limitations in the TCP reassembly code.  My intent is to
> fix this and do a better implementation (unless Jamshid is interested
> in doing it).

[catching up on old mail]

If you have started on this, I'd like to see what I could help with.
If not, I'll be looking at the other implementations soon.

I have noticed that the more recent TCP stacks set the don't fragment
flag as well.  I believe this is done to minimize the fragments...
the initial connection is used to set the maximum size which the
network can send without fragmentation.

--Michael