In order to get a better performance in distributed systems, load balancing problem has been extensively studied in recent years. Most of existing works focus on traditional systems where resources are generally homogeneous, like clusters. For grid infrastructures, this assumption is not totally true because resources of a grid are highly heterogeneous. Hence, load balancing problem for grid computing is a new challenge for scientists. In this paper, we propose a tree-based representation model for grid computing, over which we develop a hierarchical load balancing strategy. The main characteristics of this strategy can be summarized as follows:(i) It uses a task-level load balancing; (ii) It privileges local tasks transfer to reduce communication costs; (iii) It is a distributed strategy with local decision making.
Partitioning of system functionality for implementation among multiple system components, such as among hardware and software components in codesign, is becoming an increasingly important topic. Various heuristics are used in automatic partitioning. In this paper, we present our tool, called AutoDec, implemented in Visual C++ 6.0. We verified that hierarchical clustering algorithm, based on closeness metrics, can be used to merge pieces of functionality before applying Kernighan/Lin algorithm, resulting in reduced execution time with often improvements in quality. In addition, we show that our approach, when used in partitioning, fills the gap between fast algorithms and highly-optimizing ones.
