Improving MPI-HMMER’s Scalability With Parallel I/O * Rohan Darole, John Paul Walters, and Vipin Chaudhary Department of Computer Science and Engineering University at Buffalo, The State University of New York Buffalo, NY 14260 {rdarole, waltersj, vipin}@buffalo.edu May 22, 2008 Technical Report # 2008-11 Abstract As the size of biological sequence databases continues to grow, the time to search these databases has grown proportionally. This has led to many parallel implementations of common sequence analysis suites. However, it has become clear that many of these parallel sequence analysis tools do not scale well for medium to large-sized clusters. In this paper we describe an enhanced version of MPI-HMMER. We improve on MPI-HMMER’s scalability through the use of parallel I/O and a parallel file system. Our enhancements to the core HMMER search tools, hmmsearch and hmmpfam, allows for scalability through 256 nodes where MPI- HMMER was previously limited to 64 nodes. 1 Introduction As the size of biological sequence databases continue to grow exponentially, outpacing Moore’s Law, the need for highly scalable database search tools increases. Because a single processor can- not effectively cope with the massive amount of data present in today’s sequence databases newer MPI-enabled search tools have been created to reduce database search times. These distributed search tools have proven highly effective and have enabled researchers to investigate larger and more complex problems. HMMER [5, 6, 4] is perhaps the second most used sequence analysis suite. MPI-HMMER is a freely available MPI implementation of the HMMER sequence analysis suite [17, 8]. MPI- HMMER is used in thousands of research labs around the world [12, 11]. In previous work it has been shown to scale nearly linearly for small to mid-sized clusters up to 64 nodes. However, as database sizes increase, the need for greater MPI-HMMER scalability has become clear. In this paper we improve on the scalability of MPI-HMMER through the use of parallel I/O and a parallel file system. This allows us to eliminate much of the communication that previously acted * This research was supported in part by NSF IGERT grant 9987598, MEDC/Michigan Life Science Corridor, and NYSTAR. 1