An Over Time Fair Approach For Dynamic Bandwidth Allocation Of Cloud Networks

Cloud computing is becoming more and more popular and is being employed in most large data-centric information infrastructures. To fulfill flexible demands of computing power, storage and service-based computing, the pay-on-demand usage style of cloud computing shows its instinct advantage in economy. Now, data centers of various scales are being built for various requirements from enterprises, individuals and governments. Using virtualization technologies, CPU cycles, main memory and secondary storage of a physical server of a data center can be partitioned and assigned to different tenants. Meanwhile, the usages of these computing resources are isolated strictly, i.e., the computational performance on each partition (called virtual machine or VM) is guaranteed.As we know, the network performance is very important for a VM in a data center network. Data-intensive applications running in data centers, such as online multimedia services and MapReduce-like paradigms, generate a large amount of data traffic and incur massive bandwidth demands on all of the network links in the core, aggregation and edge layers of data center networks.Undoubtedly, link level bandwidth allocation policy at physical server (or hypervisor) side is one of the fundamental solutions in practice, since the hypervisor can directly monitor and manage the data traffic of contending VMs. In practice, most of global bandwidth allocation policies for data center networks should be implemented at end hosts, rather than at switches, due to expensive hardware update.Our paper is focus on the research of fair bandwidth allocation on edge level in cloud data center network. We using the historical information of the VMs’to implement the bandwidth allocation approach. The main work as follow:1) We focus on the conflicts caused by the data-intensive application between the VMs in the same physical server of the data center. And this aspect of problems is very fundamental for the hole data center network.2) We present a new definition of payment proportionality that ensures robust network sharing in inter-tenant scenarios. We also devise a bandwidth allocation policy that meets this proportionality.3) We built a real cloud platform for our proposed method. We control the bandwidth via virtual router in the cloud. Compared with the build-in congestion control policy of TCP/IP protocol and rigorous proportional share policy, the policy in our paper is both fare for the tenants’bandwidth allocation and the providers’network utilization.