Grids functionally combine globally distributed computers and information systems for creating a universal source of computing power and information. A key characteristic of grids is that resources (e.g., CPU cycles and network capacities) are shared among numerous applications, and therefore, the amount of resources available to any given application highly fluctuates over time. In this paper we analyze the impact of the fluctuations in the processing speed on the performance of grid applications. Extensive lab experiments show that the burstiness in processing speeds has a dramatic impact on the running times, which heightens the need for dynamic load balancing schemes to realize good performance. Our results

This white paper is aimed at creating a directory of existing performance monitoring and evaluation tools. The detailed categorization enables finding relevant properties, similarities and differences, and comparing the tools. The paper is neutral: there are no comments or assessment. The catalogue helps grid users, developers, and administrators in finding an appropriate tool according to their requirements. The white paper is intended to be updated by the APART community until the end of the APART-2 project and possibly beyond. Grid monitoring and performance analysis tool developers are supposed to check the categorization of their product and modify it if they find anything incorrect in our classification. Moreover, they are expected to send update messages of new versions and prototypes and the white paper will be updated accordingly. Finally, feedback about practical experiences learned from deploying and testing these tools are intended to be added to the paper in the future so that it will show a real classification among development trends.

Abstract. In this paper we describe how the CoreGrid application-level information cache mediator component will benefit from the integration of resource and service discovery mechanisms available in iGrid and Mercury. The former is a novel Grid Information Service based on the relational model. iGrid has been initially developed within the GridLab project by the ISUFI Center for Advanced Computational Technologies (CACT) at the University of Lecce, Italy. It provides fast and secure access to both static and dynamic information through a Globus Toolkit GSI (Grid Security Infrastructure) enabled web service. Besides publishing system information, iGrid also allow publication of user’s or service supplied information. The adoption of the relational model provides a flexible model for data, and the hierarchical distributed architecture provides scalability and fault tolerance. The latter, which has also been initially developed within the GridLab project by MTA SZTAKI, has been designed to satisfy requirements of grid performance monitoring: it provides monitoring data represented as metrics via both pull and push access semantics and also supports steering by controls. It supports monitoring of grid entities such as resources and applications in a generic, extensible and scalable way. It is implemented in a modular way with emphasis on simplicity, efficiency, portability and low intrusiveness on the monitored system. 1

Abstract. Monitoring systems are nowadays ubiquitous in complex environments, such as Grids. Their use is fundamental for performance evaluation, problem spotting, advanced debugging and per-use accounting. Building such systems raises challenging issues, like data gathering from Grid components, low intrusiveness, ease of use, adaptive data visualization, fault-tolerance and self-maintenance. This paper presents a new layered architecture, named Toytle, specifically designed to address these issues in the context of control Grids. All their components, from computing and network resources to complete physical processes with soft time constraints, can be monitored with Toytle. The architecture’s layers, namely the distributed core, the hierarchical connections and the local monitors, have been designed to ensure scalability, highspeed sampling and efficient dealing with large data bursts. The future Toytle implementation will adapt existing tools and also create entirely new modules.

Grids, as an emerging infrastructure for novel ways of computation, raise various theoretical and technical questions. Despite the intensive research work, these questions are sometimes even inarticulated. Is there a common understanding what grids are? Is there an accepted definition for grid performance? Are traditional performance analysis techniques adequate for grids? The paper points out problems and questions related to performance analysis in the framework of the new computing paradigm and proposes a new scenario for benchmarking.

Abstract not found