Chunk mode VM migration in XIA and triple-way pipeline for performance optimization
Introduction
For decades, Internet has been one of the most popular tools in our daily life. Developed based on TCP/IP protocol, Internet has achieved great flourish not only because of large quantity of applications, but also the various kinds of media accessing to it. Network at present is exposing more and more serious issues such as the difficulties on scalability, device mobility and security. This leads to the development and flourish of future network research.
Among the FIA projects supported by US NSF in 2010, XIA [1] complies with the concept of clean slate thoroughly, aiming at discarding TCP/IP protocol stack. It is also one of the projects included in FIA-NP (Next-Phase). Network, host, service and content can be abstracted as principals in XIA. New principal types can be defined for special use, even if the network cannot natively support the new function [2]. It is undoubtedly a good choice for incremental deployment. Network address is denoted by DAG (Directed Acyclic Graph), which is flexible for addressing. In addition, fallback allows communication entities to choose an alternative action if intent node in the DAG is unreachable. Intrinsic security is another important principle in XIA, that is, to build security considerations into the core nature of protocol design.
Virtualization is obviously a significant technology in cloud computing, which allows to run multiple operating systems on a single platform, utilizing host’s expensive resources independently. This can help to raise the working efficiency of computing resources on a physical host, especially CPU time and memory space. Data centers can achieve load balancing, host maintenance, energy management [3], network resilience [4] or disaster recovery [5] by migrating VMs.
Unfortunately, the FIA projects have not discussed much about VM migration. In this paper, we present VM migration primarily based on XIA network. Extensive research has been carried out on migration with shared-storage VMs [6], [7]. Actually, VMs with shared-storage cannot be applied in all cases. For example, a user may not have access to a particular data center permanently [8]. If a shared virtual disk is allocated, abnormal latency will occur. Therefore, we propose full migration, during which virtual disk is transferred as well as memory and CPU information. Fortunately, it is supported by block migration mechanism in QEMU-KVM hypervisor version 0.12.2 and afterward.
KVM is a piece of full virtualization hypervisor software frequently used in research area. The KVM module is integrated in the kernel of common Linux distributions. However, users cannot control Linux kernel modules directly and this issue can be solved by QEMU. QEMU is also a piece of open-source virtualization software and is adopted as a management tool in user space.
Total migration time and downtime [9] are two parameters that should primarily be taken into consideration in VM migration. Total migration time refers to the total time needed for moving a VM from source host to destination while downtime indicates the period when VM is not running and application degradation is perceived.
In this paper, we propose to study the techniques that can be used in VM migration. We attempt to accomplish VM migration base on the characteristics of XIA, rather than introducing external mechanisms. DAG format design along with routing table modification is utilized to ensure the communication of all the applications during VM migration. We design VM migration platform in two different scenarios of XIA, in a single AD (Administration Domain) and between ADs. Since VM migration in future network is quite different to that in TCP/IP network, we adopt a VM migration control protocol to manage migration procedure. We achieve virtual machine migration over XIA with the control protocol and network recovery methods. Experimental results demonstrate that virtual machine can get migrated with downtime in an acceptable range to applications and services.
The rest of this paper is organized as follows: In Sections 2 Motivation and background, 3 Related work, we discuss the motivation for this research and the related works. In Section 4, we briefly introduce the control method for VM migration which has been mentioned in previous articles and demonstrate three methods for communication recovery after VM migration. A triple-way pipeline for performance optimization is discussed in Section 5. Experimental results and VM migration performance evaluation are discussed in Section 7. Some special issues and future works are discussed in Sections 8 Conclusion, 9 Future work concludes this paper.
Section snippets
Motivation and background
Because of diversity of cloud computing platforms and resources, virtualization is necessary to maintain unified management, even for data centers in the future. Besides, virtualization is also beneficial to maintain diversity of users and applications. Therefore, virtual machine will exist for a long period of time in the future. As a typical application in future networks, it should also be well supported in XIA. There are lots of researches discussing about VM migration based on traditional
Related work
Most of recent researches on VM migration are dedicated in studying the principles of VM migration and factors that trigger migration. In general, VM migration method can be classified as pre-copy [10] and post-copy [11]. There are also optimizations based on pre-copy algorithms, such as transferring the bitmap [12] or log file [13], [14] of dirty pages, delaying “hot pages” delivery to the last round [15] to minimize the number of pages that are being transferred.
In TCP/IP network, an
Virtual machine migration in XIA
Chunk mode is introduced to conduct virtual machine migration in XIA network [25], [26]. Chunk is a connectionless method in XIA. Concept of CHUNK is widely used in content-centric future internet architecture, especially in XIA and NDN [27]. All the data transmitted are divided into chunks, which are regarded as transmission units. CID of a chunk is obtained by hash of the whole content block, so it can get self-verified. Besides, network traffic is well controlled because each transaction
Triple-way pipeline VM migration optimization
In the chunk mode VM migration control solution mentioned in Section 4 (Solution 1), the whole migration procedure is divided into chunk transmission rounds. In each round, the data sender should first calculate the CIDs of chunks and deliver these CIDs to data receiver. Simultaneously, the chunks are cached in Click of XIA. After that, the data receiver starts to request for chunks according to the CIDs and finally get the data. In this way, the network bandwidth has not been completely used
Downtime evaluation and error analysis
During a virtual machine migration procedure, the total migration time means the total amount of time needed for moving a VM from source host to destination host. The downtime indicates the period when VM is not running and application degradation is perceived.
Ping is a common but effective tool used for testing network connectivity in TCP/IP network. It can also be utilized to measure migration downtime by sending ICMP messages to VM periodically. We propose to use xping provided in XIA
Performance evaluation
In order to test the availability of our algorithm, we established two XIA testbeds, one in the lab and the other over the Internet between Tongji University and Wayne State University. The testbed topology is depicted in Fig. 12. One physical machine with multi-NIC is deployed in each campus, acting as a XIA router. The routers are running in dual-stack, that is, with both IP protocol and XIA protocol. The routers are connected by Internet and their communication is based on TCP/IP protocol.
Conclusion
This paper designs experimental testbeds in future Internet prototypes comparing against VM migration in TCP/IP network. We propose KVM virtual machines to be migrated with independent sending and receiving programs and data are transmitted in chunks. In order to solve the application interruption problem after VM is migrated, we introduce three methods for network recovery, concerning routing table modification or DAG design. All the traffics directing to VM can be recovered even if they are
Future work
This paper mainly discusses the migration control protocol to finish VM migration in XIA network, XIA islands across IP network and the testbed established based on Internet. Furthermore, we talk about how to improve the performance of chunk mode virtual machine migration in XIA network. There are still many open issues that need to be further explored.
First of all, though we achieve to implement VM migration over XIA network prototype and get some improvement on total migration time and
Acknowledgments
All the authors are sincerely grateful for the technical support from Prof. Peter Steenkiste and Principal Research Programmer Dan Barrett in Carnegie Mellon University, Prof. Weisong Shi and his students in Wayne State University.
Dalu Zhang received the B.Sc. in computer science from Shandong University, China, in 1982. He is a professor in the Department of Computer Science and Engineering, Tongji University, China. His current research interests include future Internet architecture, VM live migration, network measurement, quality of experience, semantic field, and semantic search. He has published more than 95 peer-reviewed journal articles and conference proceedings and filed 29 patents with 19 of them having been
References (27)
- et al.
A novel cache size optimization scheme based on manifold learning in content centric networking
J. Netw. Comput. Appl.
(2014) - D. Han, A. Anand, F. Dogar, B. Li, H. Lim, M. Machado, et al. XIA: Efficient support for evolvable internetworking, in:...
- A. Anand, F. Dogar, D. Han, B. Li, H. Lim, M. Machado, et al. XIA: An architecture for an evolvable and trustworthy...
- et al.
Live Migration of Virtual Machines. Symposium on Networked Systems Design & Implementation
(2005) - A. Fischer, G. Carle Fessi, H.D. Meer, Wide-area virtual machine migration as resilience mechanism, in: Reliable...
- T.S. Kang, M. Tsugawa, J. Fortes, T. Hirofuchi, Reducing the migration times of multiple VMs on WANs using a feedback...
- M. Nelson, B.H. Lim, G. Hutchins, Fast transparent migration for virtual machines, in: Proceedings of the USENIX Annual...
- S. Al-Kiswany, D. Subhraveti, P. Sarkar, et al. VMFlock: virtual machine co-migration for the cloud, in: HPDC, 2011,...
- D. Comer, A future Internet architecture that supports cloud computing, in: Proceedings of the 6th International...
- S. Akoush, R. Sohan, A. Rice, A.W. Moore, A. Hopper, Predicting the performance of virtual machine migration, in:...
Cited by (0)
Dalu Zhang received the B.Sc. in computer science from Shandong University, China, in 1982. He is a professor in the Department of Computer Science and Engineering, Tongji University, China. His current research interests include future Internet architecture, VM live migration, network measurement, quality of experience, semantic field, and semantic search. He has published more than 95 peer-reviewed journal articles and conference proceedings and filed 29 patents with 19 of them having been authorized. He is a senior member of the IEEE and ACM.
Dejiang Zhou received the B.Sc. in Nangjing University of Posts and Telecommunications, China, in 2013. He is currently a master student in Department of Computer Science at Tongji University. His research interests include future internet architecture and VM live migration.
Xiang Jin received the B.Sc. in Tianjin University, China, in 2011. He has been a student in Department of Computer Science at Tongji University and get master’s degree in 2015. His research interests mainly focus on Internet architecture and future internet architecture.