| Summary: | The minimization of tail latency is especially crucial in user interfacing services and fast responding apps. The literature on the datacenter load balancing protocols contains of many protocols but works discussing tail latency are scarce. This work proposes a novel variant of the Multi-Level Dynamic Traffic Load Balancing (MDTLB) protocol for a datacenter called the Lazy MDTLB or LMDTLB, which uses the concept of delaying the rerouting decision by a few packets for every flow that require path changes to provide the network with the time to ensure that a terrible path condition is not temporary. An evaluation of the state-of-the-art protocols of load balancing was conducted to determine the best performing one for curtailing tail latency involving flows of data mining, web search, and general flows. The findings confirmed that LMDTLB was the most efficient in minimizing tail latency and flow completion time (FCT).
|
|---|