PHYSICS

The National Center for Supercomputing Applications (NCSA) is pleased to announce the release of version 2.0 of its Virtual Machine Interface (VMI) software. VMI is a middleware communication layer that addresses the issues of availability, usability, and management in the context of large-scale SANs interconnected over wide-area computational grids. With VMI, users are able to run applications on distributed clusters that use different types of interconnects to communicate among processors. The current release supports AMD Opterons in addition to Intel's IA-64 and IA-32 platforms running the Linux operating system. The final release supports spanning a single MPI job across geographically distributed clusters. With the ability to load multiple communication devices at runtime, a high performance SAN network can be used for intra cluster communication while using TCP for inter cluster communication transparently. An MPI communication profiling framework is included for characterizing runtime communication parameters on an individual job basis to a database for subsequent analysis. MPICH-VMI supports profile guided optimization (PGO) at runtime for cross cluster communication by using the profile database. Communication characteristics from previous runs are used to optimize the mapping of virtual MPI ranks to physical processors to minimize the utilization of high latency low bandwidth links present in wide are grids. An experimental multicast communication subsystem is included with this release. Currently the MPI Broadcast collective call is implemented to utilize the multicast primitive provided by VMI. True hardware multicast is utilized if the underlying network supports it (Currently Gigabit Ethernet and Infiniband networks support hardware level multicast). Future releases will extend the spectrum of MPI collective calls utilizing multicast communication. Future releases will also include data striping across heterogeneous networks to achieve higher bandwidth than is available from a single communications interface, and dynamic failover across heterogeneous networks, a feature that will allow a computation to continue running over any remaining network interfaces if the primary interface fails. Features of VMI 2.0 include: Support for AMD Opterons and Intel's IA-64 and IA-32 platforms running the Linux operating system. Support for multiple communication interconnects, including Shared Memory for intra node communication, TCP/IP, Myricom's Myrinet GM, and Infiniband using VAPI, OpenIB and IBAL interfaces. MPI support for grid based computing with multiple communication networks active concurrently. Improved bandwidth and latency response across all communication networks. MPI latencies of 5.2 usec and peak bandwidth of 850 MB/sec have been achieved with Infiniband. A profiling interface for profile guided optimization (PGO) for large grid runs spanning high latency low bandwidth networks. An experimental implementation of MPI collectives utilizing hardware level multicast. Currently MPI Broadcast utilizes multicast communication primitives. Support for the industry standard Message Passing Interface (MPI) parallel computing API, enabling many existing codes to run in a cluster environment simply by recompiling.MPI support is based on MPICH version 1.2.5 from Argonne National Laboratory. Binary portability of MPI applications across interconnects without requiring recompilation of application software. Remote monitoring and management of the entire VMI middleware layer, allowing a user to track and modify the performance of the communications layer while using an application. Source code is available under a liberal open source-style license that allows redistribution in source or binary form provided that copyright notices and disclaimers remain intact. The release is available for download from the VMI project website (http://vmi.ncsa.uiuc.edu/). Access to documentation, a bug reporting tool, and support mail lists are also available at the site. NCSA (National Center for Supercomputing Applications) is a national high-performance computing center that develops and deploys cutting-edge computing, networking and information technologies. Located at the University of Illinois at Urbana-Champaign, NCSA is funded by the National Science Foundation. Additional support comes from the state of Illinois, the University of Illinois, private sector partners and other federal agencies. For more information, see http://www.ncsa.uiuc.edu/.