Considerations for Efficient Network Protocols and Application to Computer Graphics Clusters

Many large-scale computing applications are presently solved by a distributed system of multiple computers interconnected via a computer network [1, 12, 6]. The high availability of commodity processors and networking equipment makes construction of supercomputers from a network of commodity processors practical [15]. Many solutions for large scale visualization are based on a cluster of graphics processors connected via a commodity network [9, 10]. This makes design of efficient protocols necessary. This paper integrates ideas presented in three papers from the fields of communication theory, computer networking, and computer graphics. The three papers discuss application guided mechanisms for efficient transmission of data between multiple entities. The first paper, in the field of communication theory, is Mathematical Theory of Communication , by C. E. Shannon, published in The Bell System Technical Journal, Vol. 27, pp. 379-423, 1948 [13]. This paper presents a mathematical framework for describing communication between two hosts, in terms of information content. It also provides a mathematical encoding framework for maximizing the efficiency of transmission for noiseless and noisy channels. The second paper, in the field of computer networking, is Architectural Considerations for a New Generation of Protocols, by D. D. Clark and D. L. Tennehouse, published in Proceedings of the ACM symposium on Communications architectures and protocols, pp. 200-208, 1990 [5]. This paper introduces the principles of Application Level Framing and Integrated Layer Processing to structure new generation of protocols. It also identifies the presentation layer in computer networks as a key aspect of overall performance. The third paper, in the field of computer graphics, is WireGL: A Scalable Graphics System for Clusters , by G. Humphreys et al., published in Proceedings of 28th annual conference on Computer Graphics and interactive techniques, pp. 129-140, 2001 [10]. This paper describes the WireGL system for scalable interactive rendering on a cluster of workstations. WireGL virtualizes multiple graphics accelerators, providing a standard API for each node. It describes a formatting scheme for obtaining high rendering performance over a cluster of nodes. In addition, the authors analyze the performance of various graphics applications on clusters of varying size. In this integrative paper we summarize the three papers [13, 5, 10], we discuss the similarities and differences among them, and we attempt to integrate them. In particular, we demonstrate how the ideas of encoding from [13] are similar to the principles introduced in [5], and the application to a graphics cluster presented in [10]. The paper is organized as follows. Sections 2 and 3 describe relevant parts of the papers [13] and [5]. Section 4 highlights the common theme of these two papers and identifies the analogies and differences. Section 5 describes the paper [10]. Section 6 demonstrates the application of ideas from [13] and [5] to [10], and possible improvements. Finally our conclusions are enumerated in section 7.