首页|GCNPart: Interference-Aware Resource Partitioning Framework with Graph Convolutional Neural Networks and Deep Reinforcement Learning

GCNPart: Interference-Aware Resource Partitioning Framework with Graph Convolutional Neural Networks and Deep Reinforcement Learning

扫码查看
原文链接
  • NETL
The clouding server providers usually take workload consolidation to maximize server utilization。 For eliminating performance interference due to the competition among multiple shared resources， resource partitioning becomes an important problem in daily commercial servers scenario。 However， partitioning the critical multiple resources coordinately is particularly challenging due to the complex contention behaviors and the large search space to be explored for finding the optimal solution。 In this paper， we propose GCNPart， which focuses on allocating the optimal shared compete resource partition for colocated applications to optimize system performance。 The existing resource partitioning frameworks lack analysis and good modeling of applications， resulting in inefficiencies or lack of generality。 We formulate the resource partitioning problem as a sequential decision problem。 GCNPart builds an accurate application performance model based on graph convolutional neural networks (GCN) to learn the mapping relationships from multiple resources to applications， and then constructs deep reinforcement learning (DRL) model to consider temporal information for real-time resource partitioning decisions。 The extensive experiments evaluate that compared with the existing resource partitioning frameworks， GCNPart improves system throughput by 5。35% ～26。57%。

Workload consolidationPerformance interferenceResource partitioningDeep reinforcement learningGraph neural network

Ruobing Chen、Haosen Shi、Jinping Wu、Yusen Li、Xiaoguang Liu、Gang Wang

展开 >

NanKai University, Tianjin 300300, China

Nanyang Technological University, Singapore, Singapore

International Conference on Algorithms and Architectures for Parallel Processing

Copenhagen(DK)

Algorithms and Architectures for Parallel Processing

568-589

2022

10.1007/978-3-031-22677-9_30