Straggler-aware Observability for Flow Scheduling

Modern datacenter networks use flow scheduling to reduce flow completion time (FCT). However, schedulers without precise flow size information, like QClimb (NSDI 2024) and PIAS (NSDI 2015), can create stragglers---packets that suffer high queueing delays--undermining scheduling benefits. Current network monitoring tools find root causes of these issues but miss detecting affected "victims." In our recent work [11], we introduce StragFlow, a data plane-based system that detects stragglers in realtime at line rate and reports them to the control plane for analysis. Experiments with real-world traffic demonstrate StragFlow's effectiveness across multiple scheduling algorithms.