Reducing distributed JR program start‐up time via extending JR's operation abstraction

Abstract

SummaryThis paper shows how to simplify and speed up distributed JR program start‐up. We accomplish this goal by extending JR's operation abstraction so that it includes the creation of virtual machines and remote objects, the key components in distributed JR programs. This extension is conceptually simple, fits well with the rest of the JR language, and provides additional flexibility. Extending JR's operation abstraction yields a considerably less complex solution and a simpler implementation than other approaches, which would have led to redundant or ad‐hoc language mechanisms. The extension also allows for reduced distributed JR program start‐up costs because the creation of virtual machines and remote objects can now be done in parallel. This paper also describes how we have simulated this extension and have implemented a prototype that provides this extension. These reflect closely what we would need to do to actually include the extension within the standard JR implementation. Using the simulation and prototype, we obtained performance data on a few typical distributed applications written to use the extension. The data show large reductions, ranging from about 50% to 95%, in distributed JR program start‐up times. Our experiments also explored a few strategies to further reduce start‐up times, but, among our strategies, the straightforward strategies seem the best. This paper also discusses design alternatives for creating virtual machines and remote objects. Our experience suggests that the proposed extension to JR's operation abstraction is actually worthwhile to add to the standard JR language and implementation.