Computer Systems Organization -- Performance of Systems.

This volume contains the complete set of tutorial papers presented at the 16th IFIP (International Federation for Information Processing) Working Group 7.3 International Symposium on Computer Performance Modelling, Measurement and Evaluation, and a number of tutorial papers presented at the 1993 ACM (Association for Computing Machinery) Special Interest Group METRICS Conference on Measurement and Modeling of Computer Systems. The principal goal of the volume is to present an overview of recent results in the field of modeling and performance evaluation of computer and communication systems. The wide diversity of applications and methodologies included in the tutorials attests to the breadth and richness of current research in the area of performance modeling. The tutorials may serve to introduce a reader to an unfamiliar research area, to unify material already known, or simply to illustrate the diversity of research in the field. The extensive bibliographies guide readers to additional sources for further reading.

Multiprocessing is an effective architectural approach to enhance the performance of computer systems. However, various problems involved in multiprocessing may severely degrade system performance. This research has mainly centered on the memory interference problem in tightly coupled multiprocessor computer systems. Depending on the nature of the memory-requesting mechanism, discussion is centered on two important cases of such systems. The memory interference in multiprocessor systems with time-division-multiplexed (TDM) busses is first discussed. A general model for the memory interference in synchronous multiprocessor systems which allow arbitrary memory request rates, non-uniform memory references, and unequal processor priorities is presented next. Several application examples which make use of the memory interference models derived are presented. First, an algorithm is proposed for the estimation of the execution time of a program running in a multiprocessor system. Such an algorithm can be used to pick a computation decomposition which best utilizes the available computing power. A case study of the effect of computation decomposition on the performance of Gaussian Elimination is presented. The execution of matrix multiplication in a multiprocessor system with virtual memory was evaluated by simulation, in which a memory interference model capable of dealing with priority was used to dynamically modify various job execution times according to the number of processors and I/O channels active in the system.

As computers become more complex, the number and complexity of the tasks facing the computer architect have increased. Computer performance often depends in complex way on the design parameters and intuition that must be supplemented by performance studies to enhance design productivity. This book introduces computer architects to computer system performance models and shows how they are relatively simple, inexpensive to implement, and sufficiently accurate for most purposes. It discusses the development of performance models based on queuing theory and probability. The text also shows how they are used to provide quick approximate calculations to indicate basic performance tradeoffs and narrow the range of parameters to consider when determining system configurations. It illustrates how performance models can demonstrate how a memory system is to be configured, what the cache structure should be, and what incremental changes in cache size can have on the miss rate. A particularly deep knowledge of probability theory or any other mathematical field to understand the papers in this volume is not required.

The rapid advancement in semiconductor technology continues to change the environment in which computers are designed. As hardware costs decline, systems with multiple processors become an interesting alternative to conventional single processor system. An analytic model has been developed to describe the performance of a wide range of multiprocessor system configurations and workloads. This model deals specifically with P tightly-coupled, identical processors with shared primary and secondary memory. Secondary memory consists of a paging drum with S sectors. The workload consists of J independent, identically-distributed jobs whose faulting or I/O behavior is described by both a mean (lambda faults/sector-time) and a squared coefficient of variation (K). In addition, the processing overhead for each I/O request is added to a job's execution time at a processor (C) sector-times/fault). The model developed provides an estimate of system throughput for various numbers of processors, jobs, and drum sectors, and for various workloads. The model can be used to examine the behavior of multiprocessor systems, including the sensitivity of system throughput to each of the system parameters and parameters trade-offs related to system performance. In particular, in a system with a fixed amount of memory, the addition of jobs to the system causes a change in the memory allocation for each job and thus modifies each job's faulting behavior. The above formula for throughput is useful to examine the desirability of adding or subtracting jobs in such a system.