A parallel application benefits from scheduling policies that include a global perspective of the application's process working set. As the interactions among cooperating processes increase, mechanisms to ameliorate waiting within one or more of the processes become more important. In particular, collective operations such as barriers and reductions are extremely sensitive to even usually harmless events such as context switches among members of the process working set. For the last 18 months, we have been researching the impact of random short-lived interruptions such as timer-decrement processing and periodic daemon activity, and developing strategies to minimize their impact on large processor-count SPMD bulk-synchronous programming styles. We present a novel co-scheduling scheme for improving performance of fine-grain collective activities such as barriers and reductions, describe an implementation consisting of operating system kernel modifications and run-time system, and present a set of empirical results comparing the technique with traditional operating system scheduling. Our results indicate a speedup of over 300% on synchronizing collectives.

The problem of finding optimal reorganization intervals for linearly growing files is solved. An approximate reorganization policy, independent of file lifetime, is obtained. Both the optimum and approximate policies are compared to previously published results using a numerical example.

A statistical design of an experiment for modeling the performance (access time) of a data base (DB) system in a controlled environment has been outlined. A three-factor pilot experiment with retrieval sequence, logical view (combined with access method), and target segment type as factors was designed and the results analyzed for a specific IMS batch DB. It was found that: 1) the variation of sequences of DB retrieval calls in the application program did not have any signifilcant effect on the access time, whereas the other two factors did have significant effect; 2) the variability in access time is not completely explained by these three factors; and 3) the distribution of the residual error is "nonnormal," with a large positive skew.