UNICOS 5.1 vs. UNICOS 6.0: Performance Evaluation of Autotasking

While the efficiency of autotasking with respect to speedup values already is proved for parallel programs running in a dedicated mode, the overall system performance is of interest when parallel programs are executed in a multiprogramming environment. To evaluate the impact of autotasking on multiprogramming environments and to ensure reproducible results, there is a need to have a well-defined and constant background load. Therefore, the benchmark environment PAR-Bench was developed, generating synthetic user programs on a CRAY multiprocessor system which simulate a given workload in a flexible way by using the hardware performance monitor (HPM). Thereafter, the system can be used to run this workload several times with varied parameters; the substantial work to be done is fixed, nevertheless dynamical changes of program parameters like memory size, priority, and also variations of the degree of parallelism are supported. The PAR-Bench system provides information about mutual influences of parallel programs and background load, enabling us to evaluate different multitasking implementations, different operating systems, and different computer hardware.This paper describes our experiences with the autotasking implementations under the UNICOS operating system releases 5.1 and 6.0 in order to give an insight in the improvement of the new autotasking implementation. The results obtained so far seem to prove the efficiency of the new autotasking implementation which provides good performance in dedicated as well as in batch oriented multiprogramming environments