Damiano Zanardini

Cost analysis of java bytecode

Cost analysis of Java bytecode is complicated by its unstructured control flow, the use of an operand stack and its object-oriented programming features (like dynamic dispatching). This paper addresses these problems and develops a generic framework for the automatic cost analysis of sequential Java bytecode. Our method generates cost relations which define at compile-time the cost of programs as a function of their input data size. To the best of our knowledge, this is the first approach to the automatic cost analysis of Java bytecode.

Cost analysis of object-oriented bytecode programs

Cost analysis statically approximates the cost of programs in terms of their input data size. This paper presents, to the best of our knowledge, the first approach to the automatic cost analysis of object-oriented bytecode programs. In languages such as Java and C#, analyzing bytecode has a much wider application area than analyzing source code since the latter is often not available. Cost analysis in this context has to consider, among others, dynamic dispatch, jumps, the operand stack, and the heap. Our method takes a bytecode program and a cost model specifying the resource of interest, and generates cost relations which approximate the execution cost of the program with respect to such resource. We report on COSTA, an implementation for Java bytecode which can obtain upper bounds on cost for a large class of programs and complexity classes. Our basic techniques can be directly applied to infer cost relations for other object-oriented imperative languages, not necessarily in bytecode form.

COSTA: Design and Implementation of a Cost and Termination Analyzer for Java Bytecode

This paper describes the architecture of costa , an abstract interpretation based cos t and t ermination a nalyzer for Java bytecode. The system receives as input a bytecode program, (a choice of) a resource of interest and tries to obtain an upper bound of the resource consumption of the program. costa provides several non-trivial notions of cost, as the consumption of the heap, the number of bytecode instructions executed and the number of calls to a specific method. Additionally, costa tries to prove termination of the bytecode program which implies the boundedness of any resource consumption. Having cost and termination together is interesting, as both analyses share most of the machinery to, respectively, infer cost upper bounds and to prove that the execution length is always finite (i.e., the program terminates). We report on experimental results which show that costa can deal with programs of realistic size and complexity, including programs which use Java libraries. To the best of our knowledge, this system provides for the first time evidence that resource usage analysis can be applied to a realistic object-oriented, bytecode programming language.

Termination Analysis of Java Bytecode

Termination analysis has received considerable attention, traditionally in the context of declarative programming, and recently also for imperative languages. In existing approaches, termination is performed on source programs. However, there are many situations, including mobile code, where only the compiled code is available. In this work we present an automatic termination analysis for sequential Java Bytecode programs. Such analysis presents all of the challenges of analyzing a low-level language as well as those introduced by object-oriented languages. Interestingly, given a bytecode program, we produce a constraint logicprogram, CLP, whose termination entails termination of the bytecode program. This allows applying the large body of work in termination of CLP programs to termination of Java bytecode. A prototype analyzer is described and initial experimentation is reported.

Data dependencies and program slicing: from syntax to abstract semantics

We discuss the relation between program slicing and data dependencies. We claim that slicing can be defined, and therefore calculated, parametrically on the chosen notion of dependency, which implies a different result when building the program dependency graph. In this framework, it is possible to choose dependency in the syntactic or semantic sense, thus leading to compute possibly different, smaller slices. Moreover, the notion of abstract dependency, based on properties instead of exact data values, is investigated in its theoretical meaning. Constructive ideas are given to compute abstract dependencies on expressions, and to transform properties in order to rule out some dependencies. The application of these ideas to information flow is also discussed.

Termination and Cost Analysis with COSTA and its User Interfaces

COSTA is a static analyzer for Java bytecode which is able to infer cost and termination information for large classes of programs. The analyzer takes as input a program and a resource of interest, in the form of a cost model, and aims at obtaining an upper bound on the execution cost with respect to the resource and at proving program termination. The costa system has reached a considerable degree of maturity in that (1) it includes state-of-the-art techniques for statically estimating the resource consumption and the termination behavior of programs, plus a number of specialized techniques which are required for achieving accurate results in the context of object-oriented programs, such as handling numeric fields in value analysis; (2) it provides several nontrivial notions of cost (resource consumption) including, in addition to the number of execution steps, the amount of memory allocated in the heap or the number of calls to some user-specified method; (3) it provides several user interfaces: a classical command line, a Web interface which allows experimenting remotely with the system without the need of installing it locally, and a recently developed Eclipse plugin which facilitates the usage of the analyzer, even during the development phase; (4) it can deal with both the Standard and Micro editions of Java. In the tool demonstration, we will show that costa is able to produce meaningful results for non-trivial programs, possibly using Java libraries. Such results can then be used in many applications, including program development, resource usage certification, program optimization, etc.

Removing useless variables in cost analysis of Java bytecode

Automatic cost analysis has interesting applications in the context of verification and certification of mobile code. For instance, the code receiver can use cost information in order to decide whether to reject mobile code which has too large cost requirements in terms of computing resources (in time and/or space) or billable events (SMSs sent, bandwidth required). Existing cost analyses for a variety of languages describe the resource consumption of programs by means of Cost Equation Systems (CESs), which are similar to, but more general than recurrence equations. CESs express the cost of a program in terms of the size of its input data. In a further step, a closed form (i.e., non-recursive) solution or upper bound can sometimes be found by using existing Computer Algebra Systems (CASs), such as Maple and Mathematica. In this work, we focus on cost analysis of Java bytecode, a language which is widely used in the context of mobile code and we study the problem of identifying variables which are useless in the sense that they do not affect the execution cost and therefore can be ignored by cost analysis. We identify two classes of useless variables and propose automatic analysis techniques to detect them. The first class corresponds to stack variables that can be replaced by program variables or constant values. The second class corresponds to variables whose value is cost-irrelevant, i.e., does not affect the cost of the program. We propose an algorithm, inspired in static slicing which safely identifies cost-irrelevant variables. The benefits of eliminating useless variables are two-fold: (1) cost analysis without useless variables can be more efficient and (2) resulting CESs are more likely to be solvable by existing CASs.

Task-level analysis for a language with async/finish parallelism

The task level of a program is the maximum number of tasks that can be available (i.e., not finished nor suspended) simultaneously during its execution for any input data. Static knowledge of the task level is of utmost importance for understanding and debugging parallel programs as well as for guiding task schedulers. We present, to the best of our knowledge, the first static analysis which infers safe and precise approximations on the task level for a language with async-finish parallelism. In parallel languages, async and finish are basic constructs for, respectively, spawning tasks and waiting until they terminate. They are the core of modern, parallel, distributed languages like X10. Given a (parallel) program, our analysis returns a task-level upper bound, i.e., a function on the programs input arguments that guarantees that the task level of the program will never exceed its value along any execution. Our analysis provides a series of useful (over)-approximations, going from the total number of tasks spawned in the execution up to an accurate estimation of the task level.

Experiments in Cost Analysis of Java Bytecode

Recently, we proposed a general framework for the cost analysis of Java bytecode which can be used for measuring resource usage. This analysis generates, at compile-time, cost relations which define the cost of programs as a function of their input data size. The purpose of this paper is to assess the practicality of such cost analysis by experimentally evaluating a prototype analyzer implemented in Ciao. With this aim, we approximate the computational complexity of a set of selected benchmarks, including both well-known algorithms which have been used to evaluate existing cost analyzers in other programming paradigms, and other benchmarks which illustrate object-oriented features. In our evaluation, we first study whether the generated cost relations can be automatically solved. Our experiments show that in some cases the inferred cost relations can be automatically solved by using the Mathematica system, whereas, in other cases, some prior manipulation is required for the equations to be solvable. Moreover, we experimentally evaluated the running time of the different phases of the analysis process. Overall, we believe our experiments show that the efficiency of our cost analysis is acceptable, and that the obtained cost relations are useful in practice since, at least in our experiments, it is possible to get a closed form solution.

Resource Usage Analysis and Its Application to Resource Certification

Resource usage is one of the most important characteristics of programs. Automatically generated information about resource usage can be used in multiple ways, both during program development and deployment. In this paper we discuss and present examples on how such information is obtained in COSTA, a state of the art static analysis system. COSTA obtains safe symbolic upper bounds on the resource usage of a large class of general-purpose programs written in a mainstream programming language such as Java (bytecode). We also discuss the application of resource-usage information for code certification, whereby code not guaranteed to run within certain user-specified bounds is rejected.