DocumentCode
166994
Title
Data plane acceleration for virtual switching in data centers: NP-based approach
Author
Blaiech, Khalil ; Hamadi, Salaheddine ; Mseddi, Amina ; Cherkaoui, Omar
Author_Institution
Univ. du Quebec a Montreal, Montréal, QC, Canada
fYear
2014
fDate
8-10 Oct. 2014
Firstpage
108
Lastpage
113
Abstract
With the emerging trend of server virtualization, a new network access layer has emerged that is composed of the virtual switches running on the server platform, providing connectivity among the virtual machines (VMs) that live on the same physical server. This layer is generally implemented using the Open vSwitch (OVS) or an equivalent proprietary virtual switch. In networking for virtualization, virtual switch provides a simple solution for VM-to-VM connectivity, but this function of virtual switching must provide sustained, aggregated high-bandwidth network traffic. Majority of virtual switches implementation do not deliver adequate performance. To address this problem, we propose a strategy that aims to improve virtual switches performance by extending the packet processing tasks to hardware accelerators such as network processors. The proposed strategy is based on an adaptive and dynamic allocation of processors resources. The allocation mechanism consists on mapping the virtual switch tasks to the adequate set of resources, i.e. multi-core datapath or hardware accelerator datapath. Thus, the proposed solution tends to enhance throughput and scales down latency in order to accelerate network traffic switching over virtual switches.
Keywords
cloud computing; computer centres; multiprocessing systems; virtual machines; virtualisation; NP-based approach; Open vSwitch; VM-to-VM connectivity; adaptive allocation; data center; data plane acceleration; dynamic allocation; hardware accelerator datapath; high-bandwidth network traffic; multicore datapath; network access layer; network processor; packet processing task; server virtualization; virtual machine; virtual switching; Acceleration; Data models; Hardware; Load modeling; Program processors; Stability analysis; Throughput; Central Processing Unit; Control Theory; Network Processors; Openflow; Packet Processing; Virtualization;
fLanguage
English
Publisher
ieee
Conference_Titel
Cloud Networking (CloudNet), 2014 IEEE 3rd International Conference on
Conference_Location
Luxembourg
Type
conf
DOI
10.1109/CloudNet.2014.6968977
Filename
6968977
Link To Document