Fast dynamic process migration

Dynamic process migration supports load sharing and processor fault tolerance. We present the new freeze free algorithm for process migration, which uses six techniques to: reduce process migration latency by an order of magnitude to 19.9 ms, effectively eliminate message freeze times, and to support processor fault tolerance. The freeze free algorithm resumes execution after the transfer of four items: the combined process control and execution state, the current code page, the current heap page, and the current code page. The algorithm effectively eliminates message freeze time by separating the process state from the communication state, and thus allows message processing to proceed in parallel with process migration. The algorithm eliminates old host residual dependencies by flushing old host resident, modified data; while the process executes in parallel on the new host. The paper analyzes the costs in both the process migration latency period and the cross-network demand paging operations, and identifies further cost reductions. This paper demonstrates that small overhead is needed for good load sharing system speedup by measuring the impact of increasing overhead an speedup.