TPLCR: Time-Bound, Pre-copy Live Checkpointing and Parallel Restart of Virtual Machines Using Distributed Memory Servers

Live checkpointing of virtual machines is the ability to save the state of a virtual machine to storage while the machine is running. This paper presents a novel Time-bound, Pre-Copy Live Checkpointing and parallel Re-start mechanism (TPLCR) that implements live checkpointing based on a time-bounded, pre-copy live migration algorithm. The performance improvements of TPLCR rely on the use of multiple Distributed Memory Servers to allow fast, in-memory checkpointing and parallel restart. Along with the new TPLCR protocol, we introduce the Checkpoint-Restart Service to manage the checkpoint and restart operations in a datacenter. This paper describes a prototype implementation of TPLCR based on KVM. A series of checkpointing experiments were conducted using four CPU and memory intensive Class D NAS Parallel Benchmark kernels. Experimental results show that TPLCR checkpoint-restart performance is significantly better than traditional approaches.