Abstract. A large number of MPI implementations are currently available, each of which emphasize different aspects of high-performance computing or are intended to solve a specific research problem. The result is a myriad of incompatible MPI implementations, all of which require separate installation, and the combination of which present significant logistical challenges for end users. Building upon prior research, and influenced by experience gained from the code bases of the LAM/MPI, LA-MPI, and FT-MPI projects, Open MPI is an all-new, productionquality MPI-2 implementation that is fundamentally centered around component concepts. Open MPI provides a unique combination of novel features previously unavailable in an open-source, production-quality implementation of MPI. Its component architecture provides both a stable platform for third-party research as well as enabling the run-time composition of independent software add-ons. This paper presents a high-level overview the goals, design, and implementation of Open MPI. 1

The GridFTP extensions to the File Transfer Protocol define a general-purpose mechanism for secure, reliable, high-performance data movement. We report here on the Globus striped GridFTP framework, a set of client and server libraries designed to support the construction of data-intensive tools and applications. We describe the design of both this framework and a striped GridFTP server constructed within the framework. We show that this server is faster than other FTP servers in both single-process and striped configurations, achieving, for example, speeds of 27.3 Gbit/s memory-to-memory and 17 Gbit/s disk-to-disk over a 60 millisecond round trip time, 30 Gbit/s network. In another experiment, we show that the server can support 1800 concurrent clients without excessive load. We argue that this combination of performance and modular structure make the Globus GridFTP framework both a good foundation on which to build tools and applications, and a unique testbed for the study of innovative data management techniques and network protocols. 1

Divide-and-conquer programs are easily parallelized by letting the programmer annotate potential parallelism in the form of spawn and sync constructs. To achieve efficient program execution, the generated work load has to be balanced evenly among the available CPUs. For single cluster systems, Random Stealing (RS) is known to achieve optimal load balancing. However, RS is inefficient when applied to hierarchical wide-area systems where multiple clusters are connected via wide-area networks (WANs) with high latency and low bandwidth. In this paper, we experimentally compare RS with existing loadbalancing strategies that are believed to be efficient for multi-cluster systems, Random Pushing and two variants of Hierarchical Stealing. We demonstrate that, in practice, they obtain less than optimal results. We introduce a novel load-balancing algorithm, Clusteraware Random Stealing (CRS) which is highly efficient and easy to implement. CRS adapts itself to network conditions and job granularities, and does not require manually-tuned parameters. Although CRS sends more data across the WANs, it is faster than its competitors for 11 out of 12 test applications with various WAN configurations. It has at most 4 % overhead in run time compared to RS on a single, large cluster, even with high wide-area latencies and low wide-area bandwidths. These strong results suggest that divideand-conquer parallelism is a useful model for writing distributed supercomputing applications on hierarchical wide-area systems.

In computational grids, performance-hungry applications need to simultaneously tap the computational power of multiple, dynamically available sites. The crux of designing grid programming environments stems exactly from the dynamic availability of compute cycles: grid programming environments (a) need to be portable to run on as many sites as possible, (b) they need to be flexible to cope with different network protocols and dynamically changing groups of compute nodes, while (c) they need to provide efficient (local) communication that enables high-performance computing in the first place. Existing programming environments are either portable (Java), or they are flexible (Jini, Java RMI), or they are highly efficient (MPI). No system combines all three properties that are necessary for grid computing. In this paper, we present Ibis, a new programming environment that combines Java’s “run everywhere ” portability both with flexible treatment of dynamically available networks and processor pools, and with highly efficient, object-based communication. Ibis can transfer Java objects very efficiently by combining streaming object serialization with a zero-copy protocol. Using RMI as a simple test case, we show that Ibis outperforms existing RMI implementations, achieving up to 9 times higher throughputs with trees of objects. 1

Java offers interesting opportunities for parallel computing. In particular, Java Remote Method Invocation (RMI) provides a flexible kind of remote procedure call (RPC) that supports polymorphism. Sun’s RMI implementation achieves this kind of flexibility at the cost of a major runtime overhead. The goal of this article is to show that RMI can be implemented efficiently, while still supporting polymorphism and allowing interoperability with Java Virtual Machines (JVMs). We study a new approach for implementing RMI, using a compiler-based Java system called Manta. Manta uses a native (static) compiler instead of a just-in-time compiler. To implement RMI efficiently, Manta exploits compile-time type information for generating specialized serializers. Also, it uses an efficient RMI protocol and fast low-level communication protocols. A difficult problem with this approach is how to support polymorphism and interoperability. One of the consequences of polymorphism is that an RMI implementation must be able to download remote classes into an application during runtime. Manta solves this problem by using a dynamic bytecode compiler, which is capable of compiling and linking bytecode into a running application. To allow interoperability with JVMs, Manta also implements the Sun RMI protocol (i.e., the standard RMI protocol), in addition to its own protocol.

We present MRNet, a software-based multicast/reduction network for building scalable performance and system administration tools. MRNet supports multiple simultaneous, asynchronous collective communication operations.

One application domain the mobile computing community has not yet entered is that of grid computing -- the aggregation of networkconnected computers to form a large-scale, distributed system used to tackle complex scientific or commercial problems. In this paper we present the challenge of harvesting the increasingly widespread availability of Internet-connected wireless mobile devices such as PDAs and laptops to be beneficially used within the emerging national and global computational grid. The integration of mobile wireless consumer devices into the Grid initially seems unlikely due to the inherent limitations typical of mobile devices, such as reduced CPU performance, small secondary storage, heightened battery consumption sensitivity, and unreliable low-bandwidth communication. However, the millions of laptops and PDAs sold annually suggest that this untapped abundance should not be prematurely dismissed. Given that the benefits of combining the resources of mobile devices with the computational grid are potentially enormous, one must compensate for the inherent limitations of these devices in order to successfully utilise them in the Grid. In this paper we identify the research challenges arising from this problem and propose our vision of a potential architectural solution. We suggest a proxy-based, clustered system architecture with favourable deployment, interoperability, scalability, adaptivity, and fault-tolerance characteristics as well as an economic model to stimulate future research in this emerging field.

Previous studies of application usage show that the performance of collective communica-tions are critical for high performance computing and are often overlooked when compared to the point-to-point performance. In this paper we attempt to analyze and improve collective communication in the context of the widely deployed MPI programming paradigm by extending accepted models of point-to-point communication, such as Hockney, LogP/LogGP, and PLogP. The predictions from the models were compared to the experimentally gathered data and our findings were used to optimize the implementation of collective operations in the FT-MPI library. 1

This paper studies application performance on systems with strongly non-uniform remote memory access. In current generation NUMAs the speed difference between the slowest and fastest link in an interconnect---the "NUMA gap"---is typically less than an order of magnitude, and many conventional parallel programs achieve good performance. We study how different NUMA gaps influence application performance, up to and including typical wide-area latencies and bandwidths. We find that for gaps larger than those of current generation NUMAs, performance suffers considerably (for applications that were designed for a uniform access interconnect). For many applications, however, performance can be greatly improved with comparatively simple changes: traffic over slow links can be reduced by making communication patterns hierarchical---like the interconnect. We find that in four out of our six applications the size of the gap can be increased by an order of magnitude or more without severel...

We describe our work on improving the performance of collective communication operations in MPICH for clusters connected by switched networks. For each collective operation, we use multiple algorithms depending on the message size, with the goal of minimizing latency for short messages and minimizing bandwidth use for long messages. Although we have implemented new algorithms for all MPI (Message Passing Interface) collective operations, because of limited space we describe only the algorithms for allgather, broadcast, all-to-all, reduce-scatter, reduce, and allreduce. Performance results on a Myrinet-connected Linux cluster and an IBM SP indicate that, in all cases, the new algorithms significantly outperform the old algorithms used in MPICH on the Myrinet cluster, and, in many cases, they outperform the algorithms used in IBM’s MPI on the SP. We also explore in further detail the optimization of two of the most commonly used collective operations, allreduce and reduce, particularly for long messages and nonpower-of-two numbers of processes. The optimized algorithms for these operations perform several times better than the native algorithms on a Myrinet cluster, IBM SP, and Cray T3E. Our results indicate that to achieve the best performance for a collective communication operation, one needs to use a number of different algorithms and select the right algorithm for a particular message size and number of processes.