The constantly increasing interest devoted to grids, both in terms of developments and applications, proves that distributed computing has now attained a sufficient degree of maturity to be considered as a new network paradigm on its own. As a result, it becomes more and more unrealistic to simply …
The constantly increasing interest devoted to grids, both in terms of developments and applications, proves that distributed computing has now attained a sufficient degree of maturity to be considered as a new network paradigm on its own. As a result, it becomes more and more unrealistic to simply transpose to grid contexts, existing mechanisms that were initially thought and developed for different platforms. Grids specificities, such as the cooperating equipments number and heterogeneity, the number of independent processes, the treatments, bandwidth and stock capacities, advocate to revisit the algorithms, as well as the control and operating mechanisms, in order to reach optimal performances. But to succeed, a full comprehension of the dynamics that underly the interacting processes is a prerequisite for tailoring adapted exchange and routing strategies. The same need arose in wide area networks (Internet), that prompted the development of a dedicated metrology activity, leading researchers to resort to statistical analysis and to stochastic modeling of the traffic flows to match versatile theoretical processes. However, nothing guarantees that these models remain valid in a computing grid infrastructure, nor it is proved that traffic flows in grids present the same statistical properties as the ones evidenced with Internet.