고성능 클라우드 컴퓨팅을 위한 SCINet 기반 컨테이너 자원 관리 기법

다국어 초록 (Multilingual Abstract)

The horizontal pod autoscaling (HPA) provides scalability for cloud services composed of microservice architecture in cloud-native computing. However, the efficiency of the service is reduced and scaling is delayed due to the dynamic load occurs by various workload patterns. In addition, which is difficult to estimate the size of efficient resources for workloads composed of data of various sizes, resulting in resource waste and overloads. Therefore, this study proposes the efficient container resource management (ECoRM) that stably and elastically manages the container resources to ensure the scalability and efficiency of the service even under rapidly changing dynamic loads. The ECoRM forecasts future workloads by utilizing the sample convolutional and interaction network (SCINet) model applied with the reversible instance normalization (RevIN) method. The adaptive request generator in ECoRM generates a resource request composed of elastic size through the forecasted CPU and memory usage. Hybrid pod autoscaler scaling the resource request of pod to efficient resource sizes through ECoRM's VPA (ECo-VPA) and ECoRM's HPA (ECo-HPA), and immediately scaling the number of replicas. When the performance of ECoRM was evaluated, even if the size of resources for various workloads was incorrectly estimated, average resource utilization improved by about 17∼59% compared to the conventional HPA, and the number of overloads was measured lower than that of conventional HPA.

목차 (Table of Contents)

• 제1장 서 론 1
• 제1절 연구의 배경 및 필요성 1
• 제2절 연구의 목표 및 내용 7
• 제3절 논문의 구성 8
• 제2장 관련 연구 9
• 제1절 수직형 자원 오토스케일링 9
• 제2절 수평형 자원 오토스케일링 11
• 제3절 하이브리드 자원 오토스케일링 13
• 제3장 Efficient Container Resource Management (ECoRM) 체계 16
• 제1절 ECoRM 개요 16
• 제2절 ECoRM 구성 18
• 1. Resource metrics collector 18
• 2. Future workload forecaster 19
• 3. Adaptive request generator 25
• 4. Hybrid pod autoscaler 27
• 1) ECoRM's Vertical Pod Autoscaling (ECo-VPA) 27
• 2) ECoRM's Horizontal Pod Autoscaling (ECo-HPA) 29
• 제4장 ECoRM 설계 31
• 제5장 ECoRM 구현 34
• 제1절 Resource metrics collector 35
• 제2절 Future workload forecaster 35
• 제3절 Adaptive request generator 36
• 제4절 Hybrid pod autoscaler 39
• 1. ECo-VPA 39
• 2. ECo-HPA 40
• 제6장 ECoRM 성능 평가 41
• 제1절 Future workload forecaster 성능 평가 41
• 제2절 Hybrid pod autoscaler 성능 평가 51
• 제7장 결 론 67
• 참 고 문 헌 68
• ABSTRACT 77