Benno J. Overeinder

The distributed ASCI Supercomputer project

The Distributed ASCI Supercomputer (DAS) is a homogeneous wide-area distributed system consisting of four cluster computers at different locations. DAS has been used for research on communication software, parallel languages and programming systems, schedulers, parallel applications, and distributed applications. The paper gives a preview of the most interesting research results obtained so far in the DAS project.

Supporting internet-scale multi-agent systems

The Internet provides a large-scale environment for (intelligent) software agents. Agents are autonomous (mobile) processes, capable of communication with other agents, interaction with the world, and adaptation to changes in their environment. Current approaches to support agents are not geared for large-scale settings. The near future holds thousands of agents, hosts, messages, and migratory movements of agents. These large-scale aspects require a new approach to facilitate the development of agent applications and support. AgentScape is a scalable agent-based distributed system, described in this paper, that aims at tackling these aspects.

Scalable middleware environment for agent-based internet applications

The AgentScape middleware is designed to support deployment of agent-based applications on Internet-scale distributed systems. With the design of AgentScape, three dimensions of scalability are considered: size of distributed system, geographical distance between resources, and number of administrative domains. This paper reports on the AgentScape design requirements and decisions, its architecture, and its components.

A method for decentralized clustering in large multi-agent systems

This paper examines a method of clustering within a fully decentralized multi-agent system. Our goal is to group agents with similar objectives or data, as is done in traditional clustering. However, we add the additional constraint that agents must remain in place on a network, instead of first being collected into a centralized database. To do this we connect agents in a random network and have them search in a peer-to-peer fashion for other similar agents. We thus aim to tackle the basic clustering problem on an Internet scale and create a method by which agents themselves can be grouped, forming coalitions. In order to investigate the feasibility of a decentralized approach, this paper presents a number of simulation experiments involving agents representing two-dimensional points. A comparison between our methods clustering ability and that of the k-means clustering algorithm is presented. Generated data sets containing 2,500 to 160,000 points (agents) grouped in 25 to 1,600 clusters are examined. Results show that our decentralized agent method produces a better clustering than the centralized k-means algorithm, quickly placing 95% to 99% of points correctly. The the time required to find a clustering depends on the quality of solution required; a fairly good solution is quickly converged on, and then slowly improved. Overall, our experiments indicate that the time to find a particular quality of solution increases less than linearly with the number of agents.

The implementation of dynamite: an environment for migrating PVM tasks

Parallel programming on clusters of workstations is increasingly attractive, but dynamic load balancing is needed to make efficient use of the available resources. Dynamite provides dynamic load balancing for PVM applications running under Linux and Solaris. It supports migration of individual tasks between nodes in a manner transparent both to the application programmer and to the user, implemented entirely in user space. Dynamically linked executables are supported, as are tasks with open files and with direct PVM connections. In this paper, we describe the technical aspects of migrating message-passing tasks.

A dynamic load balancing system for parallel cluster computing

In this paper we discuss a new approach to dynamic load balancing of parallel jobs in clusters of workstations and describe the implementation into a Unix run-time environment. The efficiency of the proposed methodology is shown by means of a number of case studies.

Hierarchical resource management in the polder metacomputing initiative

A metacomputer is a set of computing resources, heterogeneous in many respects, turned into a single, uniformly accessible, computing environment. In such a system, the workload of highly demanding parallel applications that run alongside less demanding ones, has to be balanced. In the Polder Initiative we design and experiment with methods that are essential in this field. The modular structure of our metacomputer facilitates the exploration of fundamental research issues associated with load sharing and balancing, task migration, remote access, etc. In this paper we present our work in progress on a hierarchical approach in resource management (RM) for wide-area metacomputing.

Dynamite - Blasting Obstacles to Parallel Cluster Computing

Workstations make up a very large fraction of the total available computing capacity in many organisations. In order to use this capacity optimally, dynamic allocation of computing resources is needed. The Esprit project Dynamite addresses this load balancing problem through the migration of tasks in a dynamically linked parallel program. An important goal of the project is to accomplish this in a manner that is transparent both to the application programmer and to the user. As a test bed, the Pam-Crash software from ESI is used.

Integrating peer-to-peer networking and computing in the AgentScape framework

The combination of peer-to-peer networking and agent-based computing seems to be a perfect match. Agents are cooperative and communication oriented, while peer-to-peer networks typically support distributed systems in which all nodes have equal roles and responsibilities. AgentScape is a framework designed to support large-scale multi-agent systems. Pole extends this framework with peer-to-peer computing. This combination facilitates the development and deployment of new agent-based peer-to-peer applications and services.

Self-organized criticality in simulated correlated systems

In this paper we study the influence of spatio-temporal correlations on the dynamic runtime behavior of the optimistic parallel Time Warp simulation method. By means of Ising spin simulation, we show experimentally that the probability distribution of the number of rolled back events behaves as a power-law distribution over a large range of sub-critical Ising temperatures and decays exponentially for super-critical Ising temperatures. The experimental results indicate that for critical Ising temperatures, where long-range correlations occur, the computational complexity of Time Warp and physical complexity of the Ising spin model are entangled and contribute both to the runtime behavior in a nonlinear way.