CS736 Course Project Report . 1 . An In-Depth Examination of Java I/O Performance and Possible Tuning Strategies Kai Xu [email protected] Hongfei Guo [email protected] Abstract There is a growing interest in using Java for high-performance computing because of the many advantages that Java offers as a programming language. To be useful as a language for high-performance computing however, Java must not only have good support for computation, but must also be able to provide high-performance file I/O. In this paper, we first examine possible strategies for doing Java I/O. Then we design and conduct a series of performance experiments accordingly using C/C++ as a comparison group. Based on the experimental results and analysis, we reach our conclusions: Java raw I/O is slower than C/C++, since system calls in Java are more expensive; buffering improves Java I/O performance, for it reduces system calls, yet there is no big gain for larger buffer size; direct buffering is better than the Java-provided buffered I/O classes, since the user can tailor it for his own needs; increasing the operation size helps I/O performance without overheads; I/O-related system calls implemented within Java native methods are cheap, while the overhead of calling Java native methods is rather high. When the number of JNI calls is reduced properly, a performance comparable to C/C++ can be achieved. 1. Introduction There is a growing interest in using Java for high-performance computing because of the many advantages that Java offers as a programming language. To be useful as a language for high-performance computing however, Java must not only have good support for computation, but must also be able to provide high-performance file I/O, as many scientific applications have significant I/O requirements. However, while much work has been done in evaluating Java performance as a programming language, little has been done in a satisfying evaluation of Java I/O performance. In this paper, we investigate in depth the I/O capabilities of Java, and examine how and how well different possible tuning strategies work compared to C/C++. 1.1 Contribution of This Paper The contributions of this paper are threefold. First we explored possible strategies one can utilize to get high performance in Java I/O. Secondly, we designed and conducted a series of experiments that examine the performance of each individual strategy accordingly, in comparison to C/C++. Finally, experiment results are thoroughly analyzed and conclusions are reached. 1.2 Related Work There are already some papers discussing Java I/O performance. Our work is different from those in that we summarize possible I/O strategies in Java and give a thorough Java I/O performance evaluation and analysis in comparison to C/C++. [1] describes in detail possible strategies in improving Java I/O. However, no convincing experiments have been given to show how well those strategies work, neither has it studied Java I/O in comparison to that of C/C++. [2] compares Java I/O to that of C/C++ and proposes bulk I/O extensions. However, this paper mainly focuses on parallel Java I/O for specific applications instead of examining Java I/O in general.CS736 Course Project Report . 2 . 1.3 Organization The rest of this paper is organized as follows. In Section 2 we describe the basic I/O mechanisms defined in Java. In Section 3 we discuss our test methodology and experiments design. Then we give out the corresponding experiment results and analysis in Section 4. Conclusions and ideas for future work are presented in Section 5. 2. Java I/O Overview To understand the issues associated with performing I/O in Java, it is necessary to briefly review the Java I/O model. When discussing Java I/O, it is worth noting that the Java programming language assumes two distinct types of disk file organization. One is based on streams of bytes, the other on character sequences. Byte-oriented I/O includes bytes, integers, floats, doubles and so forth; text-oriented I/O includes characters and text. In the Java language a character is represented using two bytes, instead of the one byte representation in C/C++. Because of this, some translation is required to handle characters in file I/O. In this project, since our major concern is to compare Java I/O to that of C/C++, we will focus on the byte-oriented I/O. In Java, byte-oriented I/O is handled by input streams and output streams, where a stream is an ordered sequence of bytes of unknown length. Java provides a rich set of classes and methods for operating on byte input and output streams. These classes are hierarchical, and at the base of this hierarchy are the abstract classes InputStream and OutputStream. It is useful to briefly discuss this class hierarchy in order to clarify the reason why we are interested in FileInputStream/FileOutputStream, BufferedInputStream/ BufferedOutputStream, and RandomAccessFile in our test cases. Figure 2.1 provides a graphical representation of this I/O hierarchy. Note that we have not included every class that deals with byte-oriented I/O but only those classes that are pertinent to our discussion. 2.1 InputStream and OutputStream Classes The abstract classes InputStream and OutputStream are the foundation for all input and output streams. They define methods for reading/writing raw bytes from/to streams. For example, the InputStream class provides methods for reading a single byte, a byte array, or reading the available data into a particular region of a byte array. The OutputStream class provides methods for writing that are analogous to those of InputStream. OutputStream RandomAccessFile FileOutputStream FilterOutputStream BufferedOutputStream DataOutputStream InputStream FileInputStream FilterInputStream BufferedInputStream DataInputStream Figure 2.1 Pertinent Java I/O classes hierarchyCS736 Course Project Report . 3 . 2.2 File Input and Output Streams The FileInputStream and FileOutputStream classes are concrete subclasses of InputStream and OutputStream respectively, which provide a mechanism to read from and write to files sequentially. Both classes provide all the methods of their superclasses. These two classes are the lowest file I/O classes provided to