Queuing Network Approximation Method for Evaluating Performance of Computer Systems with Finite Input Source

摘要

Queuing network techniques are effective for evaluating the performance of computer systems. We discuss a queuing network technique for computer systems in finite input source. The finite number of terminals exist in the network and a job in the network moves to the server that includes CPU, I/O equipment and memory after think-time at the terminal. When the job arrives at the server, it acquires a part of memory and executes CPU and I/O processing in the server. After the job completes CPU and I/O processing, it releases the memory and goes back to its original terminal. However, when the computer system has the memory resource, the queuing network model has no product form solution and cannot be calculated the exact solutions. We proposed here an approximation queuing network technique for calculating the performance measures of computer systems with finite input source on which multiple types of jobs exist. This technique involves dividing the queuing network into two levels; one is 'inner level' in which a job executes CPU and I/O processing, and the other is 'outer level' that includes terminals and communication lines. By dividing the network into two levels, we can prevent the number of states of the network from increasing and approximately calculate the performance measures of the network. We evaluated the proposed approximation technique by using numerical experiments and clarified the characteristics of the system response time and the mean number of jobs in the inner level.