/ Omar Hammami / Associate Professor
/ Kohei Otsuyama / Assistant Professor
/ Sergey V. Ten / Assistant Professor
The Performance Evaluation Research Laboratory (PERL for short) has been studying the performance of the computer systems in terms of software and hardware interactions.
Our current major research projects are as follows:
The goal of this project is to characterize database systems, particularly in a distributed, parallel environment. Of particular interest is the instruction and data memory reference behavior. First, instruction traces of an existing database system are generated. Then, the memory reference behavior is studied and a new memory reference model is proposed. By using both original instruction traces and also the newly proposed memory reference behavior model, the memory subsystem is evaluated for the bottlenecks.
The ultimate purpose of this project is to estimate the performance of a multiprocessor systems accurately, i.e., at 1% level of accuracy. Our approach is the trace-driven simulation, which requires both an accurate hardware model at the register transfer level (RTL) and " accurate" instruction traces. The latter is particularly difficult in a multiprocessor environment, in that a trace generated in a multiprocessor environment is not necessarily a good representation in another target environment. That is, even the same workload behaves differently in different multiprocessor environment.
We have proposed a new, generative instruction behavior model for the multiprocessor environment. In order to validate the proposed model on a real computer systems, we have been trying to generate instruction traces in both priviledged and non-priviledged states of the IBM AIX operating system running in the PowerPC architecutre. We have purchased the Power PC Virtual System to generate traces. We have successfully generated a trace of a SPEC benchmark. The next step is to instrucment the AIX and generate traces of both application programs and the operating system, i.e., the AIX, simultaneously.