Fairness-improved UGAL Routing Scheme against Selfish Users in Dragonfly Networks

Abstract

Listen

Translate article

Due to the distinct advantages, including low diameter, low cost and high performance, the Dragonfly topology has revealed great potential in meeting the transmission requirements of modern data center networks. In particular, the paths in Dragonfly networks can be classified into minimal and non-minimal paths, and this path asymmetry issue makes load balancing extremely challenging. For uniform traffic pattern, minimal routing is better, while non-minimal routing is more suitable when minimal paths are congested under adversarial traffic pattern. To tackle this issue, the Universal Globally-Adaptive Load-balanced (UGAL) routing scheme, which combines both minimal routing and non-minimal routing schemes, was proposed to achieve high performance. However, due to that the selfish users prefer to select the flow with the shortest RTT (Round-Trip Time) to occupy faster paths, which results in the unfairness among users. In this paper, we develop a routing scheme called fairness-improved UGAL (F-UGAL), which aims to mitigate the unfairness in the network exacerbated by the presence of selfish behavior. F-UGAL routing decision is based on the ratio of the number of flows that have been routed to non-minimal paths to the minimal paths. Finally, the simulations show that F-UGAL can reduce the performance gained by selfish users from 59.4% to 38.9% on FCT under uniform traffic pattern, and from 43.1% to 32.3% under adversarial traffic pattern.