2019/10/08 by Cristian Hernandez Benet, Benet, Cristian Hernandez, Andreas Kassler +1 · 1 citation
Computer Science · #Software-Defined Networks and 5G #Cloud Computing and Resource Management #Interconnection Networks and Systems
paper · pdf · doi:10.48550/arxiv.1910.03324
Data center networks offer multiple disjoint paths between Top-of-Rack (ToR)\nswitches to connect server racks providing large bisection bandwidth. An\neffective load-balancing mechanism is required in order to fully utilize the\navailable capacity of the multiple paths. While packet-based load-balancing can\nachieve high utilization, it suffers from reordering. Flow-based load-balancing\nsuch as equal-cost multipath routing (ECMP) spreads traffic uniformly across\nmultiple paths leading to frequent hash collisions and suboptimal performance.\nFinally, flowlet based load-balancing such as CONGA or HULA splits flows into\nsmaller units, which are sent on different paths. Most flowlet based\nload-balancing schemes depend on a proper static setting of the flowlet gap,\nwhich decides when new flowlets are detected. While a too small gap may lead to\nreordering, a too large gap results in missed load-balancing opportunities. In\nthis paper, we propose FlowDyn, which dynamically adapts the flowlet gap to\nincrease the efficiency of the load-balancing schemes while avoiding the\nreordering problem. Using programmable data planes, FlowDyn uses active probes\ntogether with telemetry information to track path latency between different ToR\nswitches. FlowDyn calculates dynamically a suitable flowlet gap that can be\nused for flowlet based load-balancing mechanism. We evaluate FlowDyn\nextensively in simulation, showing that it achieves 3.19 times smaller flow\ncompletion time at 10% load and 1.16x at 90% load.\n