The Evolution of XTP†
Greg Chesson Silicon Graphics Computer Systems Mountain View, Ca 94039-7311 firstname.lastname@example.org
The design of XTP has been a 3 year process so far with roots traceable to work in the 1970’s and 80’s. There have been roughly ten distinct revisions of the protocol. Since Revision 3.4 in June 1989 there has been an effort to stabilize the design. This has been accomplished. The most recent versions - Revision 3.4 and 3.5 - were about 14 months apart whereas earlier versions were 3 to 6 months apart. Fortunately the net effect of the change from Revision 3.4 to 3.5 was to remove mechanism from the design. There have been at least nine distinct software implementations of the protocol. Experience with these implementations has been fed back into the protocol speciﬁcation. We believe that the functionality in the present design, Revision 3.5 plus Addendum 1a, is strengthened by the number of alternative mechanisms that have been tried and rejected. XTP history and evolution will be reviewed in this paper with a focus on some of the protocol mechanisms that have changed over time.
The quantity of network and protocol research can be measured by observing the number of publications in the area. In the 1960’s and early 1970’s most of the papers published in the biennial ACM SIGOPS Proceedings were devoted to operating system topics like paging, process control, and ﬁle management. By the 1980’s the balance changed so that most of the papers presented some aspect of networking in the context of operating systems. The papers presented in the ﬁrst workshop in a series dedicated to protocol research  reference much of the work of the 1980’s and suggest that the 1990’s may be a time of signiﬁcant progress in this area. Contemporary protocol research is inﬂuenced by developments in other technical ﬁelds. For example VLSI techniques are commonly incorporated into the design of protocols, systems, and media - something that was not possible when many standard protocols were designed. Improved media in the 100Mbit/sec to 2000Mbit/sec range limit the usefulness of protocols designed for slower media. As a result updated methods for ﬂow, rate, and error control and new services receive attention. Applications such as advanced CAD and multimedia demand greater bandwidth and low latency interfaces. As a result protocol mechanisms for providing bulk and media services, transactions, multicast, and network resource allocation become important.
† published in the Proceedings of the Third International Conference on High Speed Networking, North-Holland, 1991.
A number of writers offer insight on the current state of the art. The ﬁnal report of the ANSI X3S3.3 Study Group on High Speed Network Protocols  documents changing conditions and requirements that justify updating existing protocols or designing new ones. In a paper on emerging protocols Ian Wakeman examines several technical areas in which current protocols can be considered deﬁcient. The study published by the IBM Research Laboratory in Zurich  compares a number of protocol designs with each other and with the requirements for very high speed networking. The Xpress Transfer Protocol (XTP) is a new protocol design that has been carried out by a multidisciplinary group including a VLSI design team, an operating system team, protocol architects, and real-time system experts. There has also been feedback from several software implementations. The evolution of XTP is deﬁned by the interaction between these various interest groups over a period of years.
The chronology of XTP begins with work done at Bell Telephone Laboratories in the late 70’s and early 80’s by the author and others on lightweight protocols † for the Datakit® network - a research project that can be described as the forerunner of today’s ATM architecture. An overview of this work with...