In recent years, there has been a growing demand for the development of programs that perform analysis on large-scale graphs such as SNS and Web graphs. In evaluating the performance of such programs, we need a number of input graphs with the desired number of nodes and specific features such as the small-world property. However, the generation of large-scale random graphs by sequential programs is very time-consuming, and may cause memory shortage. In order to solve this problem, distributed parallelization of large-scale graph generation is desired.
In this research, we propose distributed parallelization of the random-graph generation based on Watts-Strogatz model, which is one of well-known graph models that provide characteristics similar to large scale graphs in the real world. We implemented our proposing distributed parallel algorithm by using Hadoop MapReduce. In the distributed parallelization, some restrictions are introduced into the original generation method to improve its efficiency. We have shown that the restrictions do not break the the characteristics that the original model has.

Embedded platforms are becoming more sophisticated as smart or IoT devices become popular. As a result, the DRAM mounted on embedded boards or chips has been increased in capacity and density to the same extent as ancient servers and PCs. The stand-alone memory diagnostic tool could be useful for detecting defective memory equipped on devices at shipping or maintenance. We aim at realizing a memory diagnostic tool using a general-purpose OS for supporting multi-platforms and for integrated diagnostic not only for memory but also for other devices. The memory diagnostic tools may have an issue that it would be hard to apply enough numbers of the test pattern into working memory on which the tool itself runs. Especially, using a general-purpose OS could impact the amount of working memory substantially. This paper proposes a tinification approach to effectively reduce the working memory to which especially OS kernel consumes by analyzing utilization and memory consumption of each kernel function. Our experimental result on an ARM platform shows 23.6% reduction in the working memory, compared with the default configured kernel.

In this paper, we propose a type system to detect non-terminating stream operations in Java programs using Stream API. Java Stream API makes it possible to write programs processing data collections in a declarative manner. Although Java Stream API has many stream operations and these operations are able to be applied to both finite and infinite streams, some operations on an infinite stream do not or may not terminate. Our proposed type system can detect these non-terminating operations. We prove the soundness of our type system and implement a type checker based on our type system with Checker Framework.

Open-source software evolves into high-performance and high-quality software by accepting submissions of source code changes from many patch authors. However, source code changes that are submitted to a project does not always obey projects' coding conventions defined. Hence, reviewers waste a large amount of time to verify source code and suggest source code improvements for patch authors. The time to improve source code can be reduced if patch authors solve these issues by themselves before patch submissions. To investigate an understandable checklist for developers in addition to coding conventions, this study analyzes source code improvement contents that are frequently solved on code review. From four Python project datasets, we found 56.0% code changes do not have an impact on behavior. We also found that only 13.4% of code improvements could be verified by using static analysis tools. As a contribution, we revealed the source code changes in code review and the improvement contents that can detect before submitting the patch by using the static analysis tools.

Much research has been done on coverage-based fuzzing tools that can generate test cases and adapt them to a testing target automatically. However, it is difficult to compare the performance of the tools because different testing targets are often used among experiments of the tools. In this paper，we performed an empirical study of four coverage-based fuzzing tools. As benchmarks, we use three collections of testing targets with their experimental setting from earlier studies. As a result, we confirmed that newer tools are statistically more effective except in a few cases.

Equations that hold on some data structures such as natural numbers or lists are called inductive theorems in equational logic. Rewriting induction is a method based on term rewriting systems developed for proving inductive theorems. Interactive theorem provers are software tools that can be used to prove mathematical theorems interactively; by mechanically applying logical inference rules, the correctness of input proofs is checked. Such activity of giving machine-checked proofs is called “formalization.” By formalizing paper proofs, it is possible to ascertain whether the proof is really correct (on the logic in consideration). Many theorems on term rewriting systems have been formalized using the interactive theorem prover Isabelle/HOL; an Isabelle/HOL library containing theorems of term rewriting systems is known as the Isabelle/HOL Formalization of Rewriting (IsaFoR). Rewriting induction, however, had not been formalized in IsaFoR. In this paper, we report a formalization of rewriting induction given in (Aoto & Toyama, 2016) in Isabelle/HOL.

This paper proposes a method to remove extraneous words in defect logs collected in software reviews to support for creating effective review checklists and scenarios used in future reviews. Extraneous words are commonly used and not specific words to capture defects to be detected in future reviews. The proposed method removes extraneous words in defect logs by excluding words in common vocabulary among defect logs collected in different types of software development projects because extraneous words are included in defect logs of any types of software development projects. This paper empirically evaluates the effectiveness of the proposed method with defect logs collected in industry. In the evaluation, words in common vocabulary among defect logs collected in reviews of a transportation system and a development support tool are excluded from those of public institutions. The results show that words contributing to creating checklists and scenarios, including “medical examination,” “fiscal year,” and “dependent in tax law,” are not excluded while extraneous words, which general stop word removal methods cannot exclude, such as “update,” “result,” and “process” are excluded.