Heap profiling of lazy functional programs

Abstract

We describe the design, implementation and use of a new kind of profiling tool that yields valuable information about the memory use of lazy functional programs. The tool has two parts: a modified functional language implementation which generates profiling information during the execution of programs, and a separate program which converts this information to graphical form. With the aid of profile graphs, one can make alterations to a functional program which dramatically reduce its space consumption. We demonstrate this in the case of a genuine example -the first to which the tool was applied -for which the results are strikingly successful. Capsule reviewProfiling technology for lazy functional programs is long overdue. Runciman and Wakeling have produced a practical and useful tool, notably because it works in the context of a fullyfledged compiler. Their work highlights the lack of information we have about the dynamic behaviour of lazy functional programs and the potential performance improvements which may be possible given profiling tools to provide us with it. This paper opens up new areas of practical research designing and building such tools.Runciman and Wakeling have started this work with the design and implementation of a heap profiling tool. They place considerable emphasis on the design of an appropriate graphical presentation of the large quantity of profiling data produced. It is interesting that they profile space and not time; ideally one would like both. An upper bound on execution time improvements from improving space behaviour is the garbage collection and paging time. However, they found that the task of fixing space bugs does sometimes lead to beneficial algorithmic changes too.We await with interest the continued development of practical profiling tools for lazy functional programs.

A Retrospective on Region-Based Memory Management

“…At this point (Summer 1995), we became aware of the work by Runciman and Wakeling on profiling of Haskell programs [46]. Based on their system, Hallenberg developed a region profiler for the ML Kit [28].…”

Profiling large-scale lazy functional programs

“…The analysis of the efficiency of a program can be addressed in two diflPerent ways Runciman and Wakeling, 1992a]. Firstly, this analysis can be done theoretically, that is by reasoning about the way in which the result is computed.…”

“…For this reason the testing of the current profiling tools found in this thesis is based purely on the effectiveness of the profiling tools for the development of a very large functional program. This work diflPers from that of Sansom which focuses on the profiling of the Haskell compiler GHC and also the work by Runciman and Wakeling [Runciman and Wakeling, 1993] in which the programmers have experience in the implementation of functional programming languages.…”

“…The problem of reasoning about the time and space complexities of programs can be addressed in two different ways; the theoretical and the practical approaches Runciman and Wakeling, 1992a . …”

Lambdas in the liftshaft---functional programming and an embedded architecture

“…It is hoped that when time profiling techniques for lazy functional languages [SPJ95] are sufficiently developed, their use will highlight further areas for improvement. Also, although space-profiling is now a well established technology [RW93], its absence from Embedded Gofer has caused certain questions about the memory requirements of applications to be left unanswered. Research into the specific patterns of memory usage found in embedded applications could eventually allow the embedded systems designer a much greater degree of control over the price/performance tradeoff.…”