Performance comparisons and data compression of time-bound live migration and pre-copy live migration of virtual machines

Pre-copy live migration is a popular virtual machine (VM) live migration mechanism that has been used in most hypervisors. In this mechanism, a key parameter, namely the maximum tolerable downtime, is required for each migration. However, defining an appropriate value for this parameter is nontrivial, especially for the migration of a virtual machine running CPU-intensive and memory-intensive applications. If an inappropriate value is used, the migration may suffer poor performance. Time-bound Live Migration (TLM) is another live migration mechanism that was proposed to solve this problem. The TLM can operate automatically without the need for the maximum tolerable downtime. Thus, the operational burden and performance penalty due to the configuration of the maximum tolerable downtime are eliminated. The contributions of this study are two folds. First, a novel compression framework for TLM, namely TLMZ, is proposed. Second, the live migration performances of TLM and TLMZ are evaluated against those of the pre-copy mechanism. In our comparisons, the pre-copy mechanism was configured to operate using a range of maximum tolerable downtime values and a number of compression options. An extensive number of experiments have been conducted. In our experiments, the TLM, TLMZ, and pre-copy mechanisms are used to migrate VMs running four OpenMP NAS parallel benchmarks. Experimental results show that TLM and TLMZ are practical solutions for the migrations of VMs running CPU-intensive and memory-intensive applications.