Study On Xen Migration Mechanism And Load Balancing Strategy

Virtualization is the key technology in cloud computing, and provides the capabilities of isolating, consolidation and migrating workload. Live migration and load balancing can improve resource utilization by allocating and adjusting system resources dynamically. However, existing studies on live migration technology and load balancing strategy do not take into account real application characteristics. As a result, both live migration and running application may experience performance degradation duo to transferring a large amount of memory pages in actual applications. Meanwhile, it leads to the poor performance of load balancing strategy.There are many approaches to improving the performance of live migration mechanism such as memory balloon, data compression, copy on write and hardware scheduling. These methods can optimize migration by reducing the amount of memory pages to be transferred. However, little research analyzed migration from the perspective of real application characteristics. On the other hand, current load balancing technologies focus on using threshold, centralized control and history data methods to produce load balancing strategy. However, these methods fail to be applied on real applications due to various kinds of applications and changing workloads. Thus, this thesis aims to develop an Adaptive Bandwidth Allocation Algorithm for Virtual Machine Migration based on Service Features, and a Load Balancing Strategy based on Complicated Service Environment. The main contributions of this thesis are as following:(1) We propose an adaptive bandwidth allocation algorithm for virtual machine migration based on service features. By calculating the memory dirty page rate of each round and analyzing the Round Trip Time (RTT) during live migration, the algorithm predicts the network bandwidth for current tasks and then allocates free bandwidth for live migration adaptively. Such method can reduce both downtime and migration time while guarantying migration performance and system reliability.(2) We propose a service-based dynamic load balancing strategy based on complicated application. The strategy can achieve resource load balancing by using extended Ganglia monitoring system to collect and distribute performance information of virtual node, extended "Black-box and Gray-box strategy" to analysis data according to the application characteristic, greedy algorithm to produce the output and Rainbow system to carry out live migration. At the same time, we add weight values in previous load balancing strategy so that the threshold can be adjusted and replace Twisted communication mechanism by Ganglia to improve the robustness of framework.(3) We implement an application-aware adaptive bandwidth allocation algorithm for live migration based on Xen. The results show our algorithm can keep QoS of application as well as guarantying migration performance by making use of free bandwidth resource reasonably. At the same time, we deploy our load balancing strategy in the real communication environment running VoIP, Streaming and C-RAN. We simulate the workloads of different services during different time period and we demonstrate the feasibility and adaptability of our load balancing strategy.