http:^^www.cs.cornell.edu^info^projects^cam^hoti-94.html

This data set contains WWW-pages collected from computer science departments of various universities

MIME-Version: 1.0
Server: CERN/3.0
Date: Monday, 25-Nov-96 00:18:59 GMT
Content-Type: text/html
Content-Length: 62736
Last-Modified: Sunday, 02-Apr-95 16:50:42 GMT

<TITLE>Low-Latency Communication over ATM Networks
using Active Messages</TITLE>
<H1><A NAME="HDT0">Low-Latency Communication over ATM Networks<BR>
using Active Messages</A></H1>

<p>
Thorsten von Eicken, Veena Avula, Anindya Basu, and Vineet Buch
<p>
Department of Computer Science<BR>
Cornell University<BR>
Ithaca, NY14850<BR>
<p>
<H2>Abstract</H2>
Recent developments in communication architectures for parallel
machines have made significant progress and reduced the communication
overheads and latencies by over an order of magnitude as compared to
earlier proposals. This paper examines whether these techniques can
carry over to clusters of workstations connected by an ATM network even
though clusters use standard operating system software, are equipped
with network interfaces optimized for stream communication, do not
allow direct protected user-level access to the network, and use
networks without reliable transmission or flow control.
<p>
In a
first part, this paper describes the differences in communication
characteristics between clusters of workstations built from standard
hardware and software components and state-of-the-art multiprocessors.
The lack of flow control and of operating system coordination affects
the communication layer design significantly and requires larger
buffers at each end than on multiprocessors. A second part evaluates a
prototype implementation of the low-latency Active Messages
communication model on a Sun workstation cluster interconnected by an
ATM network. Measurements show application-to-application latencies of
about 20 microseconds for small messages which is roughly comparable to
the Active Messages implementation on the Thinking Machines CM-5
multiprocessor.

<HR>
<H2>Table of Contents</H2>
<UL>
<A HREF="hoti-94.html#HDR1"><B>1  Introduction</B></A><BR>
<A HREF="hoti-94.html#HDR2"><B>2  Technical Issues</B></A><BR>
<A HREF="hoti-94.html#HDR7"><B>3  SSAM: a SPARCstation Active Messages
Prototype</B></A><BR>
<A HREF="hoti-94.html#HDR21"><B>4  Comparison to other approaches</B></A><BR>
<A HREF="hoti-94.html#HDR22"><B>5  Conclusions</B></A><BR>
<A HREF="hoti-94.html#HDR23"><B>6  Bibliography</B></A><BR>
</UL>

<HR>
<H2><A NAME="HDR1">1  Introduction</A></H2>
The shift from slow
broadcast-based local area networks to high bandwidth switched network
architectures is making the use of clusters of workstations<A
HREF="hoti-94.html#FN1">(1)</A> as platforms for parallel processing
more and more attractive. While a number of software packages <A
HREF="hoti-94.html#REF32249">[5</A>,<A
HREF="hoti-94.html#REF48403">6]</A> already support parallel processing
on today's workstations and networks, the communication performance is
over two orders of magnitude inferior to state-of-the art
multiprocessors<A HREF="hoti-94.html#FN2">(2)</A>. As a result, only
embarassingly parallel applications (i.e., parallel applications that
essentially never communicate) can make use of such environments.
Networking technologies such as ATM<A
HREF="hoti-94.html#REF62152">[1]</A> offer the opportunity to close the
gap: for example, ATM cells are roughly the same size as messages on
multiprocessors, it takes only a few microseconds to send or receive a
cell, ATM switches can be configured to p
From a purely technical point of view, the gap between clusters of
workstations and multiprocessors is certainly closing and the
distinction between the two types of systems is becoming blurred.