Combining explicit admission control and congestion control for predictable data transfers in grids

To improve the Grid infrastructure’s efficiency, the co-reservation of distributed resources is often required. Therefore, Grid applications need to move large amounts of data between these resources within deterministic time frames. In most cases it is possible to specify the volume and the deadline in advance. This paper proposes an approach for data-movement management and bandwidth reservation in Grid, which provides a high acceptance probability of flows in the network while maintaining efficient network-resource utilization. To achieve this, our proposal combines explicit admission control and high-speed transport protocols to enable an opportunistic sharing of the capacity by flows having heterogeneous bandwidth and delay requirements. We formulate the problem and discuss several objective functions. Then we present different heuristics and evaluate them according to the request’s acceptance rate and the network’s utilization metrics. Our simulations include all the communication and computation overheads which are involved in such data transfers.

► The mechanism provides network reservation guarantees for elastic flows in Grids.
► The heuristics consider the computation and communication overheads.
► One heuristic can accept a request even if the required bandwidth is unavailable.
► The queue-based heuristic is better when the traffic intensity is low.
► The results provide an encouraging basis for realistic admission control in Grids.