We present a new architecture for the integration of distributed resource management systems and parallel run-time environments such as MPI. The architecture solves the long-standing problem of achieving a tight integration between the two in a clean and robust manner that fully enables the functionality of both systems, including resource limit enforcement and accounting. We also present a more uniform command interface to the user, which simplifies the task of running parallel jobs and tools under a resource manager. The architecture is extensible and allows new systems to be incorporated. We describe the properties that a resource management system must have to work in this architecture, and find that these are ubiquitous in the resource management world. Using the Sun/spl trade/ Cluster Runtime Environment, we show the generality of the approach by implementing tight integrations with PBS, LSF and Sun Grid Engine software, and we demonstrate the advantages of a tight integration. No modifications or enhancements to these resource management systems were required, which is in marked contrast to ad-hoc approaches which typically require such changes.

We present a new low-latency system area network that provides the ultra-high bandwidth needed to fuse a collection of large SMP servers into a capability cluster. The network adapter exports a remote shared memory (RSM) model that supports low latency kernel bypass messaging. The Sun™ MPI library uses the RSM interface to implement a highly efficient memory-to-memory messaging protocol in which the library directly manages buffers and data structures in remote memory. This allows flexible allocation of buffer space to active connections, while avoiding resource contention that could otherwise increase latencies. We discuss the characteristics of the interconnect, describe the MPI protocols, and measure the performance of a number of MPI benchmarks. Our results include MPI inter-node bandwidths of almost 3 Gigabytes per second and MPI ping-pong latencies as low as 3.7 microseconds.

Article MPI support in the Prism programming environment Share on Authors: Steve Sistare Sun Microsystems, Inc. Sun Microsystems, Inc.View Profile , Erica Dorenkamp Sun Microsystems, Inc. Sun Microsystems, Inc.View Profile , Nick Nevin Sun Microsystems, Inc. Sun Microsystems, Inc.View Profile , Eugene Loh Sun Microsystems, Inc. Sun Microsystems, Inc.View Profile Authors Info & Claims SC '99: Proceedings of the 1999 ACM/IEEE conference on SupercomputingJanuary 1999 Pages 22–eshttps://doi.org/10.1145/331532.331554Online:01 January 1999Publication History 6citation285DownloadsMetricsTotal Citations6Total Downloads285Last 12 Months1Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

Implementors of message-passing libraries have focused on optimizing point-to-point protocols and have largely ignored the performance of collective operations. In addition, algorithms for collectives have been tuned to run well on networks of uni-processor machines, ignoring the performance that may be gained on large-scale SMP’s in wide-spread use as compute nodes. This is unfortunate, because the high backplane bandwidths and shared-memory capabilities of large SMP’s are a perfect match for the requirements of collectives. We present new algorithms for MPI collective operations that take advantage of the capabilities of fat-node SMP’s and provide models that show the characteristics of the old and new algorithms. Using the SunTM MPI library, we present results on a 64-way StarfireTM SMP and a 4-node cluster of 8-way Sun EnterpriseTM 4000 nodes that show performance improvements ranging typically from 2x to 5x for the collectives we studied.

Programs for massively parallel SIMD computers are difficult to write, understand, and debug -much more so than serial programs running on a conven­ tional computer. The barriers to understanding and debugging SIMD programs may be lowered by graph­ ically visualizing these programs as they execute. This paper describes a conceptual framework for monitoring the execution of SIMD programs in a non­ invasive fashion, and characterizes the user-interaction paradigm of a novel computer-aided design system for creating new visualizations. These ideas, and the im­ portance of visualization for SIMD programming and debugging, are illustrated in an actual debugging sce­ nario that was recorded during the use of the Inte­ grated Visualization Environment (IVE), a prototype system being developed at Harvard University and the Lockheed Palo Alto Research Labs., Advanced Com­ putation Lab.