Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!mailrus!iuvax!purdue!decwrl!ucbvax!fstohp.crd.ge.com!keith
From: keith@fstohp.crd.ge.com (Keith D Gregory)
Newsgroups: comp.protocols.tcp-ip
Subject: (none)
Message-ID: <8905121351.AA23369@crdgw1.ge.com>
Date: 12 May 89 13:53:25 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 23

Subject: UDP Connections on LOOP interface

I was wondering about the reliability of UDP connections on the LOOP interface
(127.0.0.1).  Common sense says that the packets should be delivered without
loss, unless some internal queue limit is overflowed.

I tested the basic hypothesis by sending 1.5 million packets of varying sizes
between two processes.  I then tested an internal queueing limit by creating
a socket, forking, and letting the parent sleep while the child send a small
number of packets.  Regardless of packet size, 3 packets would be queued -
any additional packets would be dropped.

So, obviously this is not a valid technique for high transmission densities.
Given low densities, however, it would be a useful replacement for msgsnd(2)
et al, especially since it would allow me to use select(2).

So, if anyone out there is able to authoritatively say yea or nay, please do.
As a note, I do not have the "real Berkeley manuals", so would be unable to
refer to such.

Thanks,
-kdg
-------