Performance Implications of Virtualization Abstract Virtualization is a component of cloud computing. Virtualization transforms traditional inflexible, complex infrastructure of individual servers, storage, and network hardware into a flexible virtual resource pool and increases IT agility, flexibility, and scalability while creating significant cost savings. Additional benefits of virtualization include, greater work mobility, increased performance and availability of resources, and automated operations. Many virtualization solutions have been implemented. There are plenty of cloud providers using different virtualization solutions to provide virtual machines (VMs) and containers, respectively. Various virtualization solutions have different performance overheads due to their various implementations of virtualization and supported features. A cloud user should understand performance overheads of different virtualization solutions and the impact on the performance caused by different virtualization features, so that it can choose appropriate virtualization solution, for the services to avoid degrading their quality of services (QoSs). In this research, we investigate the impacts of different virtualization technologies such as, container-based, and hypervisor-based virtualization as well as various virtualization features such as, over-allocation of resources, live migration, scalability, and distributed resource scheduling on the performance of various applications for instance, Cassandra NoSQL database, and a large telecommunication application. According to our results, hypervisor-based virtualization has many advantages and is more mature compare to the recently introduced container-based virtualization. However, impacts of the hypervisor- based virtualization on the performance of the applications is much higher than the container-based virtualization as well as the non-virtualized solution. The findings of this research should be of benefit to the ones who provide planning, designing, and implementing of the IT infrastructure. i Acknowledgment This research work was founded by the Swedish Foundation for Knowledge and Development, KK-Stiftelsen, in Stockholm. Part of this research is result of the collaboration between Blekinge Institute of Technology, Ericsson Company and Compuverde Company located in Karlskrona, Sweden. I would like to thank my Supervisor, Professor Lars Lundberg for his invaluable guidance and support during my studies, and for giving me this opportunity to work in a real-world industrial project. I would also like to thank my secondary supervisors David Erman and Dragos Ilie for their invaluable feedback and suggestions. In addition, I would like to thank my colleagues both in Ericsson and Compuverde for their invaluable support. I would also like to thank all my colleagues, the library staff, and the supporting departments at BTH. Finally, I am also very grateful to my family and friends for supporting, encouraging, and motivating me. ii Preface This thesis is based on the work presented in the following eight papers. The papers II, III, IV, V, and VI are published in peer-reviewed conference proceedings. Paper I is published in a journal and Paper VII is been submitted to a conference and is currently under peer-reviewing. Paper VIII is been submitted to a Journal as well. The included papers have been modified to fit this format, but the content is unchanged. Paper I S. Shirinbab, L. Lundberg, D. Erman, ”Performance Evaluation of Distributed Storage Systems for Cloud Computing”, published in International Journal of Computer Applications (IJCA), 2013. Paper II S. Shirinbab, L. Lundberg, D. Ilie, ”Performance Comparison of KVM, VMware, and XenServer using a Large Telecommunication Application”, Proceedings of the Fifth International Conference on Cloud Computing, GRIDs, and Virtualization, pp. 25-29, 2014. Paper III S. Shirinbab, L. Lundberg, “Performance Implications of Over-Allocation of Virtual CPUs”, published in International Symposium on Networks, Computers and Communications (ISNCC), pp.1-6, 2015. Paper IV S. Shirinbab, L. Lundberg, J. Håkansson, ”Comparing Automatic Load Balancing using VMware DRS with a Human Expert”, published in IEEE International Conference on Cloud Engineering Workshop (IC2EW), pp. 1-8, 2016. Paper V S. Shirinbab, L. Lundberg, “Performance Implications of Resource Over- Allocation During the Live Migration”, published in IEEE International Conference on Cloud Computing Technology and Science (CloudCom), pp. 1-6, 2016. iii Paper VI S. Shirinbab, L. Lundberg, E. Casalicchio, ”Performance Evaluation of Container and Virtual Machine Running Cassandra Workload”, published in 3rd International Conference on Cloud Computing Technologies and Applications (CloudTech), 2017. Paper VII S. Shirinbab, L. Lundberg, E. Casalicchio, ”Performance Comparison between Horizontal Scaling of Hypervisor and Container Based Virtualization using Cassandra NoSQL Database”, submitted to a conference, pp. 1-6, 2018. Paper VIII S. Shirinbab, L. Lundberg, “Scheduling Tasks with Hard Deadlines in Cloud- Based Virtualized Software Systems”, will be submitted to a Journal. iv There are other papers that are not included in this thesis but are related to this research: Paper IX E. Casalicchio, L. Lundberg, S. Shirinbab, ”Optimal Adaptation for Apache Cassandra”, published in Self Orgnizing Self Managing Clouds (SoSeMC) workshop at 13th IEE International Conference on Autonomic Computing, pp. 1-6, 2016. Paper X E. Casalicchio, L. Lundberg, S. Shirinbab, ”An Energy-Aware Adaptation Model for Big Data Platforms”, (poster) published in Autonomic Computing, pp. 1-2, 2016. Paper XI S. Shirinbab, L. Lundberg, “Real-time Scheduling in Cloud-based Virtualized Software Systems”, Proceedings of the second Nordic Symposium on Cloud Computing and Internet Technologies, pp. 54-58, 2013. Paper XII S. Shirinbab, L. Lundberg, E. Casalicchio, ”Performance Evaluation of Containers and Virtual Machines when Running Cassandra Workload Concurrently”, submitted to a Journal. Paper XIII E. Casalicchio, L. Lundberg, S. Shirinbab, ”Energy-aware auto-scaling algorithms for Cassandra virtual data centers”, published in Cluster Computing journal, vol. 20, issue 3, pp. 2065-2082, 2017. v Table of Contents 1 Introduction .................................................................................................. 1 1.1 Background .......................................................................................... 1 1.1.1 Application virtualization ............................................................. 1 1.1.2 Desktop virtualization .................................................................. 2 1.1.3 Hardware virtualization ................................................................ 2 1.1.4 Network virtualization .................................................................. 3 1.1.5 Storage virtualization.................................................................... 3 1.2 Advantages and Disadvantages of Virtualization ................................. 4 1.2.1 Advantages of virtualization ......................................................... 4 1.2.2 Disadvantages of virtualization .................................................... 5 1.3 Thesis Outline and Structure ................................................................ 6 2 Approach ...................................................................................................... 7 2.1 Related Work ........................................................................................ 7 2.2 Aim and Scope ................................................................................... 10 2.3 Research Questions ............................................................................ 10 2.3.1 Research question 1 .................................................................... 11 2.3.2 Research question 2 .................................................................... 11 2.3.3 Research question 3 .................................................................... 11 2.3.4 Research question 4 .................................................................... 12 2.3.5 Research question 5 .................................................................... 12 2.3.6 Research question 6 .................................................................... 12 2.3.7 Research question 7 .................................................................... 13 2.3.8 Research question 8 .................................................................... 13 2.4 Research Methodology ....................................................................... 14 2.4.1 Experimental Study .................................................................... 14 2.4.2 Theoretical and Simulation Study .............................................. 15 3 Results ........................................................................................................ 15 vi 3.1 Contributions ...................................................................................... 15 3.1.1 Contributions in Paper I.............................................................. 15 3.1.2 Contributions in Paper II ............................................................ 15 3.1.3 Contributions in Paper III ........................................................... 16 3.1.4 Contributions in Paper IV ........................................................... 16 3.1.5 Contributions in Paper V ...........................................................