Chill: Runtime Reconfiguration of Windowed Stream Aggregates with Semantic Guarantees

Paper in proceeding, 2026

Stream Aggregates are a key building block in edge-to-cloud data pipelines used to summarize data over windows, time intervals defined by their advance (the frequency of output) and size (the interval length). In heterogeneous deployments such as smart grids and vehicular networks, runtime changes in workload and resources require practitioners to rebalance freshness, cost, and semantic guarantees.A limitation of existing solutions is that, once the window advance and size are chosen, they stay fixed for the entire execution. To overcome this limitation, Chill enables dynamic adjustment of both parameters. Analysts can specify a set of acceptable advances and sizes, selecting them at runtime to trade performance and guarantees through weaker (at-most-once) or stronger (exactly-once) reconfiguration semantics. We present alternatives for both time stamp-ordered and out-of-order execution models. We also empirically show that Chill can match the performance of fixed-configuration baselines while supporting reconfigurations in negligible time.