Comparison of hash strategies for flow-based load balancing

Abstract

Hashing is powerful tool and widely used for flow-based load balancing schemes in parallel processing. In this paper, we analyze and compare computing overhead and load dispersion characteristics of hash strategies using XOR and CRC operations under four hashing key schemes (from 2-tuple to 5-tuple). We conduct experiments with real-life 24-hour campus network traffic. The results show that XOR32 has the lowest computing overhead among all hash function groups. Moreover, XOR32 with 4-tuple and XOR32 with 5-tuple are the two outstanding strategies that provide very good uniform distribution of traffic across multiple links, thus achieving better load balancing for flow-based applications.