COMMUNICATION MECHANISMS FOR THE DISTRIBUTED COMPUTING ENVIRONMENT

Abstract

Communication is perhaps the most widely discussed issue in the field of Distributed Computing Systems (DCS). Communication is essential to facilitate the co-ordination among processes in a DCS. There are many communication mechanisms available for distributed computing. A review and analysis of these communication mechanisms are presented in chapter 2. A single communication mechanism cannot cater to all kinds of distributed applications. As such, a communication architecture that provides an integrated, comprehensive and coherent view of several useful communication mechanisms is required. However, most of the existing distributed systems only concentrate on defining distributed system architecture, and provide only one communication mechanism for distributed computing. On the other hand, although the OSI model defined by ISO can be used as a communication architecture for distributed computing, it is not specially designed for that purpose; instead its main objective has been to provide a general reference model for interconnecting heterogeneous computer systems. This thesis discusses a new architectural communication model for distributed computing. The proposed new model, called SHILPA is a layered communication architecture that incorporates several useful communication mechanisms. SHILPA provides a rich set of communication mechanisms to the application programmer and thus facilitates the building of diverse distributed applications. The communication mechanisms defined include an asynchronous remote procedure call mechanism (ASTRA), a Type-safe Message-Passing mechanism (TMP), a Sun eXternal Data Representation (XDR) protocol, a Byte-stream oriented Message-Passing mechanism (BMP), and a Reliable Datagram Transport Protocol (RDTP). Each of these communication mechanisms forms part of a layered architecture with consistent interfaces between them. SHILPA is described in detail in chapter 3. The highest layer in SHILPA is an asynchronous RPC mechanism (ASTRA). Synchronous RPC is a useful communication mechanism for distributed computing. However, it fails to exploit the parallelism inherent in a distributed system because the caller blocks while waiting for the callee to complete. To remedy this shortcoming in the synchronous RPC model, ASTRA was developed. ASTRA calls do not the block the client, thus allowing the client execution to proceed in parallel with the server invocation. The result is buffered at the client side when it is returned by the server, and the client can claim the results later. ASTRA also allows a client to specify explicitly whether a call is to be optimized for low-latency or high throughput. Moreover, ASTRA optimizes any intra-machine RPC calls by bypassing expensive network and data conversion operations. A more detailed discussion of ASTRA is presented in chapter 4. Chapter 5 introduces TMP, XDR and BMP. TMP and BMP are the message-passing communication mechanisms provided by SHILPA, and XDR is the data presentation protocol selected in SHILPA. BMP is the lowest asynchronous message-passing layer in SHILPA. It consists of four classes of services: intra-machine, virtual circuit, datagram and reliable datagram. Type conversion is not provided in BMP; data exchanged are purely in byte stream format. TMP is an asynchronous message-passing mechanism. It provides type-safe data transfer between heterogeneous machines. There are three sets of communication services in TMP, namely virtual-circuit, datagram and reliable datagram. TMP uses the XDR and BMP services to provide type-safe message passing. XDR is needed in the SHILPA architecture to ensure that data are correctly represented in different machines. RDTP was developed to provide reliable delivery of messages-in the datagram format. RDTP provides a reliable end-to-end transport of data between host processes in a connectionless and orderly manner. It is particularly useful for building distributed client-server applications which are largely involved in intermittent exchanges due to its low-cost, low latency and connectionless nature. Chapter 6 provides a comprehensive description of the RDTP protocol.