schmitz@fas.ri.cmu.edu (Donald Schmitz) (11/28/89)
For at least the last year, I've heard rumors of a real-time system benchmark in the works, called Rhealstone (sp?), however I've never seen any references or results. Does anyone know if the benchmark project is for real, still under development, etc? If not, what are YOU using for benchmarking systems? Seems lots of people use task switch times, but this doesn't tell you much about the scheduler performance, or even overall utilization, since it is easy to strip the context switch code down by moving the scheduling/sorting somewhere else. If you don't have any formal benchmarks, how about an informal poll as to typical job mixes (I'll collect mailed responses and summarize if there is interest). In particular, for your application, what are: o number, rate, and duration of periodic tasks o acceptable clock skew in task execution o long and short term utilization o number, rate, latency and computation requirements of interrupts and their handlers o number of active processes o number of exclusion protected resources and duration of locked accesses o number (and type) of processor As you may have guessed, we are working on our own real-time kernel, and wonder how well some of our design decisions work in real applications, as opposed to handling pathological cases or yielding blazing fast context switch times. All responses greatly appreciated. Don Schmitz (schmitz@fas.ri.cmu.edu)