Refine
Document Type
Language
- English (2)
Has Fulltext
- no (2)
Is part of the Bibliography
- no (2)
Keywords
- Application Code (1)
- Co-Allocation (1)
- Code Parameter (1)
- Component Technology (1)
- Distributed Systems (1)
- Executable Code (1)
- Globus Toolkit (1)
- Grid Application (1)
- Higher-Order Components (1)
- Resource Management (1)
Institute
This paper deals with one of the fundamental properties of grid computing – transferring code between grid nodes and executing it remotely on heterogeneous hosts. Contemporary middleware relies for this purpose on Web Services, which makes application programs complicated and low-level and requires much additional expertise from programmers. We compare two mechanisms for grid application programming with regard to their handling of code transfer – the de-facto standard WS-GRAM in Globus and the higher-level approach based on HOCs (Higher-Order Components). We study the advantages and problems of each approach using a real-world application case study – the sequent alignment problem from bioinformatics. Our experiments show the trade-off between reduced development costs and software complexity when HOCs are used and the higher performance of the applications on the grid when using WS-GRAM.
Grid applications are increasingly being developed as workflows using well-structured, reusable components. We argue that components with well-defined semantics facilitate an efficient scheduling on the Grid. We have previously developed a user-transparent scheduling approach for Higher-Order Components (HOCs) – parallel implementations of typical programming patterns, accessible and customizable via Web services. Our approach combines three scheduling techniques: using cost functions for reducing communication overhead, reusability of schedules for similar workflows, and the aggregated submission of jobs. We analyze the user-transparent scheduling from four perspectives, namely: the easiness of integration within already existing Grid scheduling systems, the gains for individual users, the resource provider advantages, and the robustness with respect to execution failures. We perform our evaluation using the KOALA Grid scheduler extended to support our user-transparent scheduling, which we run on the DAS-2 system combining over 200 nodes at five sites in the Netherlands. The experimental results show an increase in throughput by more than 100%, a descreasing of the response time by 50%, and a failure reduction by 45% for the considered scenarios