The ParaStation2 Cluster Environment

The key to efficient parallel computing on workstations clusters is a communication subsystem that removes the operating system from the communication path and eliminates all unnecessary protocol overhead. But the key to effective and widespread cluster computing in general is to provide a cluster environment which offers high performance and well known programming interfaces as well as an integrated management and administration of the cluster that makes it suitable for a widespread community. The ParaStation2 cluster environment fulfills these requirements. Its one-way latency for all programming interfaces is about (depending on the hardware platform) and throughput is 100 to 150 MByte/s. It offers standard programming interfaces, including PVM, MPI, Unix sockets, Java sockets, and Java RMI which allow parallel applications to be ported to ParaStation2 with minimal effort. ParaStation’s one system image is responsible for management and administration of the cluster in terms of fault tolerance (automatic removal and reintegration of faulty nodes), node management (offering logical nodes rather than physical machines), load balancing (automatic mapping of appropriate nodes to applications), and job control and disaster management in case of faulting applications. The system is implemented on a variety of platforms (Alpha workstations running Tru64 Unix, as well as Intel and Alpha’s running Linux) and is proven to operate large clus-