An implementation of a concurrent gang scheduler for a PC-based cluster system

• Published in: Systems and computers in Japan 2007-03, Vol.38 (3), p.39-48
• Main Authors: Hyoudou, Kazuki, Kozakai, Yasuyuki, Nakayama, Yasuichi
• Format: Article
• Language: English
• Subjects: cluster-based computing; gang scheduler; IEEE 1394; parallel processing systems
• ISSN: 0882-1666; 1520-684X
• DOI: 10.1002/scj.20458

In recent years PC‐based cluster systems have been used more and more due to their excellent cost performance and high degree of flexibility. With a view to constructing systems at lower costs and with sufficient performance capacity, we have designed and implemented FireCluster that consists of a PC‐based cluster system in which nodes are connected via the IEEE 1394 interface. Parallel job scheduling is important to use modern multiprocessor parallel computers efficiently. For example, gang scheduling is suitable for cluster systems that provide a user‐level communication library and many gang schedulers have been implemented for such systems. However, on low‐cost cluster systems with relatively low communication performance, the communications costs associated with a gang scheduling system whereby all nodes are synchronized each time the system switches jobs can be a significant problem. Therefore, we focus on concurrent gang scheduling and have implemented a concurrent gang scheduler for PC‐based cluster systems. The experimental results show that the system is able to perform synchronization with low communications costs, switch jobs with a low overhead, and is highly scalable to increases in the number of nodes. © 2007 Wiley Periodicals, Inc. Syst Comp Jpn, 38(3): 39–48, 2007; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20458