Flow Diffusion Algorithms for Folded Clos Networks

抄録

The performance of a data center network is a key to provide high quality information services. As a topology for data center networks, a folded Clos network (FCN) is advantageous because it provides high throughput for arbitrary traffic patterns. To maximize the throughput of an FCN, traffic congestion must be avoided. This is achieved by uniformly diffusing flows between links. Thus, this paper investigates two distributed algorithms that nearly uniformly spread flows in FCNs. The first algorithm is executable with information that is locally obtainable at the source of a flow. This method is further divided into two versions. One version may reroute an existing flow for a flow completion, while the other does not perform rerouting. The second algorithm runs with semi-local information, which can be obtained via communication between the source and destination switches. These algorithms are executable in a distributed manner without relying on global information. This means that the processing overhead for these algorithms is small. This paper theoretically derives upper bounds on the number of flows that go through a link when these algorithms are applied. These bounds assure that the load offered on a link does not grow extremely heavy. This paper also evaluates the presented algorithms and conventional random routing for different traffic and network models through computer simulation. The results clearly show the advantage of the presented algorithms.