在虛擬資料中心處理時變工作量的資源管理方法

Resource Orchestration Techniques for Time Varied Workload in Virtualized Datacenters

薩亞吉(Padhy, Satyajit)

指導教授 ： 周志遠

國立清華大學/電機資訊學院/資訊系統與應用研究所/博士(2021年)

摘要

現代虛擬化數據中心通常依靠虛擬機（VM）遷移將工作負載整合到一台機器上以節省能源。但是虛擬機遷移有很多缺點，包括性能下降、服務中斷等。因此，當遷移發生時，使用整合感知調度方法來最小化開銷是微不足道的。如果工作負載隨時間變化，即隨時間變化的工作負載，則手頭的問題變得更具挑戰性。隨著工作負載需求的變化，需要一個編排器來相應地管理資源。在本論文中，我為虛擬化數據中心中隨時間變化的工作負載提出了各種資源編排技術。我已經在兩種不同類型的系統上實現了這種方法：(i) 批處理系統和 (ii) 網絡功能虛擬化 (NFV)。 我將論文分為三部分進行研究：（i）CAMIRA是一種可感知合併的調度算法，可主動避免VM遷移；（ii）MIRAGE是一種使用遺傳算法將VM遷移減至最少的VM放置策略 (iii)REAP 是一種重新配置感知編排技術，適用於虛擬化數據中心中具有時變工作負載的網絡功能虛擬化。在CAMIRA中，我們的結果表明，利用VM整合決策的先驗知識，我們的調度算法將遷移數量顯著減少了37％-46％，從而提高了系統性能，同時減少了網絡帶寬消耗並縮短了作業執行時間。在 MIRAGE 中，我們展示了同樣的問題可以表述為整數線性規劃（ILP）問題，並且可以通過遺傳算法找到接近最優的解決方案。 在 REAP 中，我們研究了各種重新配置策略，通過在 NFV 佈局問題中設計模型來製定其成本，ILP 和啟發式解決方案。在我們的評估中，我們觀察到能源和重新配置成本降低了 60% 和 25%，我們分析了每種重新配置方法的成本降低，以驗證我們啟發式算法的設計。

關鍵字

分佈式計算 ； 調度算法 ； 優化問題 ； 資源管理 ； 算法設計 ； 網絡功能虛擬化

並列摘要

Modern virtualized data centers often rely on reconfiguration solutions like migrations and scaling to address huge workload demands or consolidate workload for energy saving. But reconfiguration solutions has many drawbacks, for instance VM migrations can cause performance degradation, service disruption etc. Hence, it is trivial to use reconfiguration-aware scheduling approach to minimize the overhead when migrations occur. The problem in hand becomes more challenging if the workload varies across time, i.e. time-varied workload. As the workload demand changes, an orchestrator is required to manage resources accordingly. In this thesis I have proposed various resource orchestration techniques for time-varied workload in virtualized datacenters. My methodologies are focused on two different types of systems: (i) batch system and (ii) Network Function Virtualization (NFV). And I have divided my thesis into three different works: (i) extbf{CAMIRA} is a consolidation aware scheduling algorithm that proactively avoids VM migrations, (ii) extbf{MIRAGE} is a VM placement strategy that uses a genetic algorithm to minimize VM migrations and (iii) extbf{REAP} is a reconfiguration aware orchestration technique for NFV with time-varied workload in virtualized datacenters. In CAMIRA our results show that by taking advantage of the prior knowledge of VM consolidation decision, our scheduling algorithm significantly reduces the number of migrations by 37\% - 46\% which leads to improved system performance with less network bandwidth consumption and shorter job execution time. In MIRAGE we show the same problem can be formulated as an integer linear programming~(ILP) problem, and an near-optimal solution can be found by using a genetic algorithm. In REAP we investigate various reconfiguration strategies for different problem domain in NFV, formulate their costs by designing a model in the NFV placement problem, and propose ILP and heuristic solutions. In our evaluation, we observed 60\% and 25\% cost reduction for energy and reconfiguration costs, and we analyze the cost reduction from each of the reconfiguration methods to validate the design of our heuristic algorithm.

並列關鍵字

Distributed Computing ； Scheduling Algorithm ； Optimization Problem ； Resource Management ； Algorithm Design ； Network Function Virtualization

參考文獻

[1] S. Akoush, R. Sohan, A. Rice, A.W. Moore, and A. Hopper. Predicting the performance of virtual machine migration. In IEEE International Symposium on Modeling, Analysis Simulation of Computer and Telecommunication Systems, pages 37–46, 2010.

[2] Paul Barham, Boris Dragovic, Keir Fraser, Steven Hand, Tim Harris, Alex Ho, Rolf Neugebauer, Ian Pratt, and Andrew Warfield. Xen and the art of virtualization. In Proceedings of the Nineteenth ACM Symposium on Operating Systems Principles, SOSP ’03, pages 164–177, New York, NY, USA, 2003. ACM.

[3] M.F.Bari, S.R.Chowdhury, R.Ahmed, andR.Boutaba. Onorchestrating virtual network functions. In 2015 11th International Conference on Network and Service Management (CNSM), pages 50–56, Nov 2015.

[4] M.F.Bari, S.R.Chowdhury, R.Ahmed, andR.Boutaba. Onorchestrating virtual network functions. In 2015 11th International Conference on Network and Service Management (CNSM), pages 50–56, 2015.

[5] L. A. Barroso and U. Hölzle. The case forenergy-proportionalcomputing. Computer, 40(12):33–37, Dec 2007.