A buddy scheme for combining load sharing and fault tolerance in a distributed computing system is proposed and evaluated. The results show that this scheme can achieve the expected both load sharing and fault tolerance at the cost of one.

This is a conference paper version of the technical report on EMUNET.

Issues of synchronization and rollback and their relations in distributed simulations are discussed. Frequent synchronization results in more communications overhead, but reduces the possibility of rollback.

TCP/IP functions are replaced by simulated module so that debugging can be done in memory. The advantage is that the behave of the distributed programs can be repeated easily.

The feasibility and the performance of two distributed simulation schemes in a loosely coupled multi-computer environment supported by the TCP/IP protocol are investigated and reported. Executing distributed simulation on a uni-processor computer degrades performance significantly.

Event links are used to restore or cancel incorrectly scheduled events. This technique greatly reduces the space requirement to support distributed simulation while causing very little performance degradation in time.

Empirically study shows that a class of simulation problems make use of the optimistic distributed simulation very efficiently. Event links may be used to cancel wrongly scheduled events, instead of saving detailed copies of events.

This is a simulation study of the combined load sharing and fault tolerance scheme. The simulation results confirms our analytical model developed earlier.

An approximation for the mean end-to-end packet delay is presented for a ring-based, two-level MAN with asymmetric traffic. An asymmetric polling model based on Boxma and Meister's work is used. Simulations show that the approximation is very accurate over a wide range of network parameters. No on-line version is available.

A simulation model for a ring-based, two-level MAN is presented. The study shows that the average service time (packet length) has more impact on the average delay than the average arrival rate in the case of asymmetric traffic. No on-line version is available.

A ring-based, two-level MAN is modeled as a single-server, multi-queue system with non-exhaustive cyclic service and non-zero switchover times. An approximate solution was obtained using Boxma and Meister's one-level, non-exhaustive, asymmetric polling model. No on-line version is available.

Traditional Mean Value Analysis (MVA) requires that the network satisfy product form. This study presents an approximate MVA algorithm that can handle difference classes of customers with varying service requirement at a FCFS service center. No on-line version is available.

This program demonstrates sorting algorithm using graphics capability of the IBM PCs. The executable can be run on any IBM 8086, 80286, 80386 and 80486 machines. The on-line file is in pkzip format. The files included there are the executable, the fonts file, and the source code which is in Turbo-Pascal 1.0

A solution to a truncated Markov chain is discussed. Truncated Markov chain can be used in many technical applications. The results were compared against the exact results in M/M/1 to prove that the solution is valid. A method of automatically generating a system of equations is also proposed.

The effect on the end-to-end delay of the number of local networks, the capacity ratio of the local networks and the backbone network are examined. Both analytical and simulation model are used in the study. No on-line version is available.