INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 02, FEBRUARY 2020 ISSN 2277-8616 452 IJSTR©2020 www.ijstr.org Data Race Detection Techniques In Java: A Comparative Study Devesh Lowe, Dr. Mithilesh Kumar Dubey, Bhavna Galhotra Abstract: Most modern programming languages support multi -threading, which is commonly required in executing parallel programs in mobile multi - media applications, web-servers and operating systems. Concurrency bugs are common in multi-threaded applications. Bugs like race conditions, data races and deadlocks are some problems which arise out of unsynchronized code but have to capacity to sneak in to large and complex structures and may result in giving unexpected results. Data Race is one such synchronization bug, arising out of unsynchronized code, which can stay hidden even after many rounds of testing, but can emerge as a system destroyer when it surfaces. Researchers, in last four decades have presented various tools to detect data races which are broadly classified into static, dynamic and hybrid categories. In this paper authors have focussed on data races arising in object oriented programming language- java and have presented a study of various research articles subjected towards early detection of data races. Authors hereby present a study of techniques used by different researchers and have tried to summarise the java oriented approaches to solve the problem. Authors also present a comparison on different java based tools used for data race detection. Index Terms:Parallel Processing, Concurrency Bug, Race Condition, Dynamic Data Race Detection, Java ——————————  —————————— 1. INTRODUCTION Multi-threaded programs are the main stay of modern programming. Most modern programming languages support multithreaded applications, which definitely improves the responsiveness of programs. Use of multi- threaded applications give rise to concurrency bugs like race conditions, data races and deadlocks. Data race is one of the most common concurrency bug that is affecting multithreaded programming since its inception. A data race arises when two or more threads try to access a shared resource and one of them is in write mode [1] i.e. trying to change or update the shared resource. Data race is notoriously difficult to detect as ordering of thread execution is decided by operating system where program or programmer don’t get any access. A data race if not handled properly has the potential to even crash a system. All parallel programming languages provide mechanisms which help to prevent data races. It is important that a data race caused by an imperfect code must be detected early to prevent them from causing much harm to the system. In many situations, data races and race conditions are often mistaken for each other. There is a peculiar causal and non- causal relationship between Race Conditions and Data Races. Many race conditions are caused by data races and similarly many data races are caused by race conditions. There are also situations where data races are independent of race conditions and vice-versa. Race Condition occurs when timing or ordering of events gets affects by other undesirable or uncontrollable events resulting in hampered programs correctness. For example situations created by context switch, OS signals, hardware interrupts and memory operations on multi-processors can produce race conditions. Data race is more of a programming error of synchronization. As explained by Sebastian Burckhardt at Microsoft Research, Data Race happens when there are two memory accesses in a program where both: [2]  Target the same location  Are performed concurrently by two threads  Are not reads  Are not synchronization operations Various data race detection techniques have been developed in last four decades which have been instrumental in early detection of this anomaly. Researchers have categorised these detection techniques as on-the-fly, ahead of time and post mortem techniques [3]. Broadly classified as Static and Dynamic techniques. Static data race detection techniques or ahead of time techniques use compile time heuristics. On-the-fly approaches are dynamic in nature. Considering the frequency of occurrences of data races in a multi-threaded application, most studies have focused upon dynamic detection of data races keeping a check over the number of false positives as minimum. We observed that race detectors or tools developed by various researchers were primarily enhancements over previously written tools with an attempt to minimize false positives. Since data races are more frequent in multi -threaded ————————————————  Devesh Lowe is currently pursuing Phd and is an assistant professor, Jagan Institute of Management Studies, Rohini, New Delhi, India, 110085,devesh.lowe@jimsindia.org  Mithilesh Kumar is an Associate Professor, Lovely Professional University Punjab, India, mithilesh.21436@lpu.co.in  Bhavna Galhotra is pursuing Phd and is an assistant professor, Jagan Institute of Management Studies, Rohini, New Delhi, India, 110085, bhawna.galhotra@jimsindia.org