Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!cs.utexas.edu!rice!sun-spots-request
From: tai@cs.ucla.edu (Ann Tai)
Newsgroups: comp.sys.sun
Subject: User process network lantency
Keywords: Networks
Message-ID: <7373@brazos.Rice.edu>
Date: 4 May 90 23:05:39 GMT
Sender: news@rice.edu
Organization: Sun-Spots
Lines: 11
Approved: Sun-Spots@rice.edu
X-Sun-Spots-Digest: Volume 9, Issue 146, message 13

We are implementing a two-node SPARCstation system which requires the
message transmission latency to be minimum.  We implemented the software
for inter-machine communication using stream socket and datagram socket.
Our experience is that the latency for sending a message between two SUN
SPARCstation 1 (connected by 10 megabits/sec ethernet) is greater than 150
milliseconds using datagram and is worse using scream socket.  Literatures
showed that user process network latency, defined as the minimum time
required to send a single byte of data, can be below 5 milliseconds. If
you know the right way for us to measure the latency and how to achieve
the best response time (latency), please give us your suggestion.  Thank
you in advance.