الفهرس | Only 14 pages are availabe for public view |
Abstract The classic way to measure the performance of a system is to simulate the behavior of that system when executing a typical workload.The workload used in the simulation is produced by tracing the execution of that workload on existing real computer and the execution trace subsequently drives a simulation of the design variations of interest.Synthetic traces on the other hand have several advantages.They are faster than real traces and may be just as useful for evaluation purposes.Synthetic traces may be the only possible approach for gathering traces for a new architecture for which no typical workloads exist.The error in an evaluation caused by using a synthetic trace need not be any larger than the errors produced in the selection of a representative workload by selecting a specific real trace from many possible traces.Synthetic workload is controlled and can be used to stress a system under designSmall storage size is needed to store the trace generator program instead of storing the whole trace file which facilities sharing of traces which was very difficult in the past.Generating synthetic traces is also very cheap no extral resources are needed,simple uniform and hyperbolic random number generators are used to generate the trace. |