Performance based cluster architecture: Analytical modeling and analysis
Date of Completion
Fundamental to the development and use of parallel and distributed systems is the ability to observe, analyze and understand their performance at different levels for different application types and across alternative system and software environments. First we used the Product Form Solution (PFS) of Jackson networks to construct a steady state model for evaluating the performance of clusters of workstations. The model includes the major parameters that determine the cluster performance. It covers the effect of storage limitations, interconnection networks and the impact of data partitioning. The model can be used to estimate the steady state throughput of the cluster or the expected service time of the tasks under any specific configuration and different workloads. It can also detect the bottlenecks in the system, which can then be improved leading to more effective utilization of the available resources. ^ Jackson networks have been very successful in modeling computer systems. However, the ability of Jackson networks to predict performance with system changes remains an open question, since they do not apply to systems where there are populations size constraints. ^ Also, the PFS of Jackson networks assumes steady state and exponential service centers or certain specialized queueing disciplines. To overcome these limitations, we present a more general transient model for Jackson networks that is applicable to any population size, service distribution and any finite workload (no new arrivals). It reduces to Jackson networks where Jackson networks are applicable. Using several non-exponential distributions, we show to what extent the exponential distribution can be used to approximate other distributions and transient systems with finite workloads. We study the cases where the non-exponential servers have no contention (PFS still applies), and also where non-exponential servers have queueing (Jackson networks can't be applied). The model can be considered as the base line for the cluster architecture analysis because it evaluates the cluster behavior without using any special task or scheduling algorithms. It can also be used to investigate efficient techniques to data allocation, resource management and dynamic scheduling. ^
Mohamed, Ahmed Mostafa Abdel-Rahman, "Performance based cluster architecture: Analytical modeling and analysis" (2004). Doctoral Dissertations. AAI3144600.