Skip to Main Content
Memory efficiency is important for processing high-volume data streams. Previous stream-aggregation methods can exhibit excessive memory overhead in the presence of skewed data distributions. Further, data skew is a common feature of massive data streams. The authors introduce the AdaptWID algorithm, which uses adaptive processing to cope with time-varying data skew. AdaptWID models the memory usage of alternative aggregation algorithms and selects between them at runtime on a group-by-group basis. The authors' experimental study using the NiagaraST stream system verifies that the adaptive algorithm improves memory usage while maintaining execution cost and latency comparable to existing implementations.