William R. Franta

Analysis of a prioritized CSMA protocol based on staggered delays

SummaryCarrier sense multiple access (CSMA) protocols were originally conceived for use in digital radio communication. Such protocols are, however, quite applicable in a variety of distributed computing and local networking configurations based on a shared bus. The CSMA protocols are characterized by the fact that message collisions can occur only in a very small time window whose duration corresponds to the propagation delay of the bus. They are further characterized by the technique used to subsequently serialize the transmission of the colliding messages. In this paper we analyze the situation wherein colliding messages are serialized by using a set of staggered ‘retry’ delays which are associated with the sources that may generate the colliding messages. Our analysis produces both path throughput (utilization) curves and message delay curves, as a function of message generation rate. These curves show that the protocol is in a sense self stabalizing, serve to quantify its behaviour, and indicate loading levels which result in unacceptable message delays.

An efficient data structure for the simulation event set

Recently algorithms have been presented for the realization of event scheduling routines suitable for general purpose discrete event simulation systems. Several exhibited a performance superior to that of commonly used simple linked list algorithms. In this paper a new event scheduling algorithm is presented which improves on two aspects of the best of the previously published algorithms. First, the new algorithms performance is quite insensitive to skewed distributions, and second, its worst-case complexity is O(√n), where n is the number of events in the set. Furthermore, tests conducted to estimate the average complexity showed it to be nearly independent of n.

An efficient collision-free protocol for prioritized access-control of cable or radio channels☆

Abstract We describe and analyze an efficient collision-free channel-access protocol for cable or radio networks with an arbitrary spatial (one-, two- or three-dimensional) fixed node configuration. The protocol is robust and provides distributed access-control under a myriad of possible priority disciplines, including Fixed, Fair Round-Robin and Prioritized Round-Robin disciplines. The protocol optimally employs available information on network topology, to provide performance characteristics (throughput and delay) that are at least as good as and in most cases much better than those of other published protocols that employ less information. The performance advantage is especially significant for networks with a large number of nodes.

Message-based priority access to local networks

Hyperchannel interfaces are designed to interface computer mainframes and peripheral equipment to a high speed bus to form a local network. Each interface executes a distributed control contention access algorithm operation revolves about fixed priorities assigned to the individual interfaces. The access algorithm is extended to provide for message-assigned priorities, to provide better bus access rights for higher classes of messages at the expense of degraded access rights for the lower classes of messages. The extension requires little additional new hardware but instead employs a replication in each interface of existing hardware. The original hyperchannel interface access mechanism is explained. The algorithm extension is described and two models are developed to assess its effect on message transmission access rights, and via these models the viability of the extended algorithm can be demonstrated quantitatively.

A process oriented simulation model specification and documentation language

We present a language for specifying and documenting discrete event simulation models, designed to facilitate communication during model design, implementation and maintenance. The systems from which models are developed are viewed as collections of communicating elements. Each element is specified separately in a module consisting of attributes, axioms and a scenario. The models specification consists of each model elements specification and some systemwide attributes and axioms. Synchronization among model elements (model dynamics) is specified within the model elements rather than globally. Each language control structure has a monitor associated with it. Appropriate changes are made in the models state in response to a monitors sensing an attribute change. The language contains a library of specifications for common simulation model entities such as the priority queue. The syntax and semantics of the language are presented through examples. Examples of a library specification and a model specification are included.

A comparison of heaps and the TL structure for the simulation event set

Following publication of our paper [2], questions arose with respect to the superiority of the TL structure over heaps,1 particularly in the face of the remarks of Gonnet [3], concerning the use of heaps for the physical realization of the simulation event set. Gonnets communication was in response to the Vaucher and Duval paper [5], and suggested the heap to be a more efficient structure than any proposed in [5]. As regards a comparison of heaps and the TL structure we can make the following remarks:

Real-Time Distributed Computer Systems

Publisher Summary This chapter describes real-time distributed computer systems in regard to their design and implementation. It discusses the motivation for distributed computer systems in terms of possible system characteristics attained by distributing the computational resources. The chapter also characterizes the real-time control application environment. Further, the chapter reviews the options and issues related to hardware and software designs for distributed systems, and examines the details of the design and implementation of the Honeywell Experimental Distributed Computing System (HXDP). Distributed computer systems essentially contain several computers and provide increased system availability and reliability. Their design is complex, involving the design of communications mechanisms in hardware and software, and the selection of policies and mechanisms for distributed system control. However, the complex design issues can have simple solutions in a well-understood application environment. Distributed systems are considered attractive as they provide efficiency, modularity, robustness to failure, and extensibility. Moreover, for real-time environments, these systems allow physical distribution of functionality, placing computing power at places where required.

A diffusion approximate solution to the g/g/k queueing system

Abstract In the opening sections of this paper we present a diffusion approximate solution to a G/G/k queueing station under steady state condtions. The approximation technique is computationally simple and practical in that it requires knowledge of the means and variances of the interarrival and service time distributions, but not their distributional forms. Several numeric examples attesting to the good quality of the approximation are also included. We then discuss extending the basic approximation technique to account first for a feed back option and then to networks of multiserver queueing stations.

A generalized simulator for computer networks

A generalized network simulator has been constructed to help investigate the performance of computer networks at a variety of levels. The simulator accounts for user-host, host-node (a node is a communications interface unit), and node-node interactions. The program is highly modular so that the user can change one level without affecting the others; this structure allows for innovation and efficient experimentation. Model validation has been a major consideration in the develop ment of the simulator. We have employed a formal procedure to validate models at the input-output level by comparing them to simple lumped models for a given experimental frame. The simulator can handle a variety of network structures and protocols. A particular network is described in terms of a schedul ing function, a grouping function, and a distance control matrix. Several network topologies and protocols have been simulated successfully. One problem with the modular approach is that it may lead to inefficient programs. Our simulator, for example, had to perform many scans of its events list because of the existence of numerous simultaneous events. We removed this inefficiency by introduc ing a new event scheduling routine based on a secondary key. The secondary key assigns priorities to simultaneous events.

The mathematical analysis of the computer system modeled as a two stage cyclic queue

SummaryThe use of network queueing models to aid in the analysis and design of multiprogrammed computer system is well established. Their use provides qualitative insights as well as quantitative measures. These latter measures may be used either directly to assess system performance, or indirectly to provide a theoretical backstop for the validation of more elaborate simulation modes. Further, we note that the mathematical analysis of these models is of itself interesting and worthy of consideration.In this paper we concentrate on the simplest possible network model, that is, that model which consists of two nodes, through which jobs cyclically and repeatedly pass in their bid for service from the node servers. As we shall hopefully demonstrate a wide variety of situations can be represented by variations of this simple model. In this endeavor we shall include a consideration of several variants which, at least to date, defy exact mathematical analysis. In these latter cases we are therefore much interested in the availability of good quality approximate solutions, and several are discussed.As our title and preceding remarks suggest it is the mathematical analysis of these models on which we concentrate, pointing out for each, of course, what its analysis contributes to the totality of information.In the closing section we review our effort and point out the relationship of our results to more general network models.