This paper introduces Madeleine II, a new adaptive and portable multiprotocol communication library. Madeleine II has the ability to control multiple network protocols (BIP, SISCI, VIA) and multiple network adapters (Ethernet, Myrinet, SCI) within the same application session. Moreover, it includes advanced mechanisms to dynamically select the most appropriate transfer method for a given network protocol according to various parameters such as data size or responsiveness user requirements. We report on performance measurements obtained using BIP and SCI and we present preliminary results about our NEXUS/Madeleine II and MPICH/Madeleine II ports.

This paper introduces Madeleine II, an adaptive and portable multi-protocol communication library for high performance multithreaded applications. Madeleine II has the ability to control multiple network protocols (BIP, SISCI, VIA) and multiple network adapters (ETHERNET, MYRINET, SCI). Moreover, it includes advanced mechanisms to dynamically select the most appropriate transfer method for a given network protocol according to various parameters such as data size or responsiveness user requirements. We report on performance measurements obtained using various protocols and we present preliminary results about porting the MPICH and the NEXUS communication libraries on top of Madeleine II.

In this document we make a brief review of memory management and DMA considerations in case of common SCI hardware and the Virtual Interface Architecture. On this basis we expose our ideas for an improved memory management of a hardware combining the positive characteristics of both basic technologies in order to get one completely new design rather than simply adding one to the other. The described memory management concept provides the opportunity of a real zero-- copy transfer for Send--Receive operations by keeping full flexibility and efficiency of a nodes' local memory management system. From the resulting hardware we expect a very good system throughput for message passing applications even if they are using a wide range of message sizes. 1 Motivation and Introduction PCI--SCI bridges (Scalable Coherent Interface [12]) become a more and more preferable technological choice in the growing market of Cluster Computing based on non--proprietary hardware. Although absolut...

Although many ideas about protected user-level network access have been around for a while there still hasn't been a satisfying solution to combine extreme low-latency and highbandwidth over a wide range of message sizes. Uniting the Scalable Coherent Interface (SCI), a technology to build distributed shared memory as well as message passing systems on top of it, and the Virtual Interface Architecture (VIA), a generic architecture aiming at low latency, seems to be a proper solution to tackle the problem. To reap the benefits of this approach we are developing a new Message Passing Interface (MPI) library, called CHEMPI 1 . In this paper we give a brief introduction into some relevant design principles of CHEMPI and prove its potential regarding latency and bandwidth on base of a comparative measurement including two other computing interface technologies. Keywords: MPI, VIA, SCI, shared memory, and communication protocols. 1

The availability of a comprehensive software infrastructure is essential for the success of a parallel architecture. In order to allow for the greatest possible flexibility, an infrastructure has to be designed in an integrated, easy-to-use manner and with the support of multiple parallel programming paradigms and models to address a wide code base. Shared Memory in a LAN-like Environment (SMiLE) provides such an infrastructure for SCI (Scalable Coherent Interface) based clusters. It includes support for a large range of message passing libraries as well as for almost arbitrary shared memory programming models. In addition, SMiLE contains initial work on an appropriate tool set for performance optimization. The complete infrastructure is closely optimized for the underlying hardware and therefore offers its benefits to the user without significant overheads.

This document will give you an impression of what MP-MPICH is and then help you to install and use the different subsystems of MP-MPICH

Normally the SISCI interface provides a Distributed Shared Memory (DSM) [PTM97] abstraction using the Scalable Coherent Interface (SCI). This paper describes and discusses the design and the concepts behind a library called VIA2SISCI that we have developed. The library maps the semantics of the Virtual Interface Architecture (VIA) to SISCI--semantics and establishes a middleware between SISCI and high-- level communication facilities. We focus on several important concepts of VIA that had to be mapped to the SISCI services. The presented concepts may be interesting and useful beyond the scope of this paper.

Abstract. This paper presents communication strategies for supporting efficient non-blocking Java communication on clusters. The communication performance is critical for the overall cluster performance. It is possible to use non-blocking communications to reduce the communication overhead. Previous efforts to efficiently support non-blocking communication in Java have led to the introduction of the Java NIO API. Although the Java NIO package addresses scalability issues by providing select() like functionality, it lacks support for high speed interconnects. To solve this issue, this paper introduces a non-blocking communication library to efficiently support specialized communication hardware. This library focuses on reducing the startup communication time, avoiding unnecessary copying, and overlapping computation with communication. This project provides the basis for a Java Message-passing library to be implemented on top of it. Towards the end, this paper evaluates the proposed approach on a Scalable Coherent Interface (SCI) and Gigabit Ethernet (GbE) testbed cluster. Experimental results show that the proposed library reduces the communication overhead and increases computation and communication overlapping. 1

Abstract. Since version 2 of the Message Passing Interface (MPI) the MPI job has the ability to start new processes dynamically at runtime. The MPI implementation has to care for the creation of new communication channels in those situations. This work desribes the implications of dynamic process creation on an communication module for the Scalable Coherent Interface (SCI). The module is developed for the PTL framework of Open MPI. 1

Handling groups of processes is a desirable characteristic in many parallel and distributed applications, being used for providing performance and availability. In this paper we present the group communication service that is being developed for the DECK programming environment, showing how it was implemented and analyzing its preliminar performance. Keywords: group communication, distributed environments, cluster computing. 1 Introduction For a great number of parallel and distributed applications, groups of processes are frequently needed to provide collective communication facilities in order to allow a set of distributed tasks to exchange data. This kind of facility is supported by communication group mechanisms that offer resources for an application to control a set of processes that are concurrently running within a group context. DECK (Distributed Execution and Communication Kernel) [BAR98, BAR00] is a library for parallel programming on clusters of PCs that foresees the av...