RE: IPCOMP and IPSEC

[Paul Koning <[e-mail]>]

>>>>> "Avram" == Avram Shacham <[e-mail]> writes:

 Avram> At 12:01 PM 6/4/98 -0400, Robert Moskowitz wrote:
 >> TCPng needs to add intelligent compression (that is interact with
 >> the application).  There is could have history.

 Avram> In previous discussions of compression at level 4, several
 Avram> people correctly pointed that TCP-compression may reduce the
 Avram> number of IP packets while IP compression can only reduce the
 Avram> size of each packet. Fewer IP packets may enhance performance
 Avram> even more than utilizing compression history.

That's a very good point.  Certainly sending the same number of bytes
in a smaller number of packets is always a win.

 Avram> But - and this may be a BIG implementation obstacle - the
 Avram> current compression algorithms require ~16KB of compression
 Avram> and decompression context for _each_ connection.  In other
 Avram> words, 16KB per socket...

That's only a few millicents per socket.

 Avram> Also, UDP is still a useful L4 protocol and no stateful
 Avram> compression is possible here either.

True, but for UDP packets that are larger than the MTU, the same
IP packet reduction benefit applies.