Optimizing marshalling by run-time program generation

  • Authors:
  • Barış Aktemur;Joel Jones;Samuel Kamin;Lars Clausen

  • Affiliations:
  • University of Illinois at Urbana-Champaign;University of Alabama;University of Illinois at Urbana-Champaign;State's Library, Aarhus, Denmark

  • Venue:
  • GPCE'05 Proceedings of the 4th international conference on Generative Programming and Component Engineering
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Saving the internal data of an application in an external form is called marshalling. A generic marshaller is difficult to optimize because the format of the data that will be marshalled is unknown at the time the marshaller is implemented. On the other hand, efficient marshallers can be written for specific kinds of data. In this paper we use run-time program generation (RTPG) to produce specialized marshallers. We use Jumbo, a Java compiler supporting programmer-specified RTPG. We show that RTPG is easily employable. Speedups in order of magnitude can be achieved in some cases. We study the case where the data consist of a large number of objects of a single class and the case where there are objects of many classes. In the latter case, “just-in-time” heuristics allow us to limit RTPG costs and gain considerable speedups.