next up previous contents
Next: Related Work Up: Performance Previous: Simulation-Based Evaluation

Overall System Performance

We do not have the ideal measurements to show the benefit of non-blocking synchronization for overall system performance. However, in other work [25], system performance has been shown to benefit considerably from the ability to execute code in signal handlers as exploited extensively by the Cache Kernel object-oriented remote procedure call system. This system allows restricted procedures, namely those that do not block, to be executed directly as part of the signal handler invocation that handles a new call. With this optimization, many performance-critical RPCs can be invoked directly in the signal handler without the overhead of allocating and dispatching a separate thread to execute the RPC. Our measurements, reported in the cited paper, indicate a significant savings from this optimization, particularly for short-execution calls that are common to operating system services and simulations.



Michael Greenwald
Thu Sep 19 12:18:20 PDT 1996