Hierarchical Autoscaling for Large Language Model Serving with Chiron

TL;DR

This paper introduces Chiron, a hierarchical autoscaler for large language model serving that improves resource utilization and SLO adherence by considering request-specific SLOs and employing backpressure estimation.

Contribution

Chiron is a novel autoscaling approach that incorporates request SLOs and hierarchical backpressure to optimize LLM serving efficiency.

Findings

Chiron achieves up to 90% higher SLO attainment.

Chiron improves GPU efficiency by up to 70%.

Outperforms existing autoscaling solutions in LLM serving.

Abstract

Large language model (LLM) serving is becoming an increasingly important workload for cloud providers. Based on performance SLO requirements, LLM inference requests can be divided into (a) interactive requests that have tight SLOs in the order of seconds, and (b) batch requests that have relaxed SLO in the order of minutes to hours. These SLOs can degrade based on the arrival rates, multiplexing, and configuration parameters, thus necessitating the use of resource autoscaling on serving instances and their batch sizes. However, previous autoscalers for LLM serving do not consider request SLOs leading to unnecessary scaling and resource under-utilization. To address these limitations, we introduce Chiron, an autoscaler that uses the idea of hierarchical backpressure estimated using queue size, utilization, and SLOs. Our experiments show that Chiron achieves up to 90% higher SLO attainment and improves GPU efficiency by up to 70% compared to existing solutions.