Optimizing Live Migration of Virtual Machines in SMP Clusters for HPC Applications

A Small-Scale Testbed for Large-Scale Reliable Computing

“…Improving the reliability of high performance computing (HPC) systems is one of the leading research areas in the HPC field. Many studies have been performed that have covered various methods for understanding how, why, and when failures occur in large-scale HPC systems (Atif & Strazdins, 2009;DeBardeleben, Blanchard, Fu, Guan, & Zhang, 2011;Fu & Xu, 2007;Hacker et al, 2009;Oliner & Stearley, 2007;Pandit, Kalbarczyk, & Iyer, 2009;Romero, 2010;Salfner & Tschirpke, 2008;Zhang, Squillante, Sivasubramaniam, & Sahoo, 2004;Zheng, Lan, Park, & Geist, 2009;Zhou, Zhan, Meng, Xu, & Zhang, 2010;.…”

“…The live migration of virtual machines is a process which involves copying the running virtual machine's memory pages to another physical host also running a VM and, in the case of Xen, iteratively copies the memory pages up to 30 times before pausing the running operation of the VM, resuming the VM on the new host, and then destroying the original VM instance (Atif & Strazdins, 2009). Atif and Strazdins (2009) hypothesized that the overall performance of live migrations could be significantly improved by reducing the amount of CPU intensive and network intensive memory copy iterations to the bare minimum of two iterations. Atif and Strazdins (2009) demonstrated that optimizing this process resulted in a reduction of memory page transfer by over 500% for the very memory intensive HPC benchmarks.…”

“…Atif and Strazdins (2009) hypothesized that the overall performance of live migrations could be significantly improved by reducing the amount of CPU intensive and network intensive memory copy iterations to the bare minimum of two iterations. Atif and Strazdins (2009) demonstrated that optimizing this process resulted in a reduction of memory page transfer by over 500% for the very memory intensive HPC benchmarks. Their optimization showed improvements of almost 50% in total wall clock time spent in the live migration process and around 200% performance improvement for memory intensive and network intensive jobs (Atif & Strazdins, 2009).…”

“…Atif and Strazdins (2009) demonstrated that optimizing this process resulted in a reduction of memory page transfer by over 500% for the very memory intensive HPC benchmarks. Their optimization showed improvements of almost 50% in total wall clock time spent in the live migration process and around 200% performance improvement for memory intensive and network intensive jobs (Atif & Strazdins, 2009). Additionally, live migration has already been shown to be a promising proactive fault tolerance technique.…”

Challenges and New Solutions for Live Migration of Virtual Machines in Cloud Computing Environments

“…When it exceeds 90%, the migration process moves to the stop-and-copy phase. Atif et al [6] propose a more rigid termination strategy. According to their statistics on Xen's default migration mechanism for HPC applications, iteratively copying the memory data of HPC applications is only a waste of time and CPU cycles.…”