Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!uunet!brunix!euclid!freedman
From: freedman@euclid.math.temple.edu (Avi Freedman)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: UDP - How unreliable?
Message-ID: <22899@brunix.UUCP>
Date: 14 Dec 89 12:04:25 GMT
References: <1039@anasaz.UUCP>
Sender: news@brunix.UUCP
Reply-To: freedman@euclid.math.temple.edu (Avi Freedman)
Organization: Math Department, Temple University, Philadelphia, PA
Lines: 26

In my experience with UDP staying on one side of a DEC bridge and
going between a Sun 4/280 and 5 Sun 3/60s, I lost roughly one out
of 10,000 UDP packets.  However, this is not at full Ethernet load-
I think that the load will be a _major_ factor in how many UDP packets
get lost.  Also, a slow machine (or a fast machine with a slow
ethernet card) may lose packets, and if you're going through
gateways/bridges/routers/whatever, they could hiccup or slow down
or experience high network loads and lose packets.

You get the idea.  For applications like the one I was working on
(simulating data collection of massive quantities of data points
and computing statistics about them), it was not critical if a packet
or two got lost, since the entire message was always contained in
one UDP packet.

Of course, for other applications, where order is dependant or duplication
would be unacceptable (implementing a database that sits on the
network, for example), I just use TCP.

Sorry to be so vague, but UDP packet-loss depends on many factors.  If
you'd like, I'll send you code to determine how many UDP packets are 
lost...


				Avi Freedman
				freedman@euclid.math.temple.edu