Publication | Closed Access
Enable simultaneous DNN services based on deterministic operator overlap and precise latency prediction
57
Citations
28
References
2021
Year
Unknown Venue
Deterministic Operator OverlapEngineeringMachine LearningPrecise Latency PredictionComputer ArchitectureLatency PredictabilityLatency UnpredictabilitySystems EngineeringParallel ComputingPerformance ImprovementPerformance PredictionComputer EngineeringScheduling (Computing)Computer ScienceDeep LearningRuntime SystemSignal ProcessingReal-time SystemsParallel Programming
While user-facing services experience diurnal load patterns, co-locating services improve hardware utilization. Prior work on co-locating services on GPUs run queries sequentially, as the latencies of the queries are neither stable nor predictable when running simultaneously. The input sensitiveness and the non-deterministic operator overlap are two primary factors of the latency unpredictability. Hence, We propose Abacus, a runtime system that runs multiple services simultaneously. Abacus enables deterministic operator overlap to enforce latency predictability. Abacus composes of an overlap-aware latency predictor, a headroom-based query controller, and segmental model executors. The predictor predicts the latencies of the deterministic operator overlap. The controller determines the appropriate operator overlap for the QoS guarantee of all the services. The executors run the operators as needed to support the deterministic operator overlap. Our evaluation shows that Abacus reduces 51.3% of the QoS violation and improves the throughput by 29.8% on average compared with state-of-the-art solutions.
| Year | Citations | |
|---|---|---|
Page 1
Page 1