A statistical mechanical approach to a framework for modeling irregular programs on distributed or cluster computers

Abstract

Correctly modeling the resource requirements of a computer program is made problematic by the property of irregularity. Irregular computer programs are ones in which modeling the program before runtime is neither feasible nor possible; the result being that the systems requirements are unknown. We present a method for modeling irregularity which makes use of well established probabilistic and data analysis techniques. The realm of continuous mathematics, rather than the more traditional discrete cases are employed, thus introducing the more diverse analysis methodologies this branch of mathematics affords. This approach, that of describing a discrete system using a continuous mathematical function has been well established in the physical sciences, and has proven to be a valid approach to describing problems of this sort.