Local-area networks based on high-bandwidth packet-switching technology, such as ATM, show a tremendous promise in reducing communication latencies and overheads. However, the lack of flow-control and reliable delivery in ATM networks requires the higher protocol layers to deal with cell loss or corruption. While it is possible to use TCP-based communications over ATM, the protocol mismatch results in a significant loss of the bandwidth. In addition to this, there is also a significant mismatch between the stream-based communication model provided by TCP, and the message-oriented communication model required by the parallel applications.This paper describes the design and implementation of a reliable messaging layer aimed at workstation clusters using broadcast and packet-switched local area networks.
[1]
Seth Copen Goldstein,et al.
Active messages: a mechanism for integrating communication and computation
,
1998,
ISCA '98.
[2]
Henry M. Levy,et al.
Efficient Support for Multicomputing on ATM Networks
,
1993
.
[3]
Guang R. Gao,et al.
FTL: a multithreaded environment for parallel computation
,
1994,
CASCON.
[4]
David E. Culler,et al.
A case for NOW (networks of workstation)
,
1995,
PODC '95.
[5]
Thorsten von Eicken,et al.
Low-latency communication over ATM networks using active messages
,
1994,
Symposium Record Hot Interconnects II.
[6]
Ramón Cáceres.
Measurements of Wide Area Internet TraffiC
,
1989
.
[7]
David D. Clark,et al.
Architectural considerations for a new generation of protocols
,
1990,
SIGCOMM '90.