Concurrent Checkpoint/Restart Mechanism for Real-Time Interactive Processes

This paper presents a new concurrent checkpoint/restart mechanism, it allows the checkpointed process (checkpointee) without stopping while doing checkpointing to some extent. The checkpointee can continue to run until a write request is captured by tracing TLB during dumping memory pages (which is a critical step of doing checkpoint). At that time, the checkpoint module will copy the requested page to the designated memory buffer, then enable the write operation. Then, it can form an image file with consistent process state by using original pages which copied in the designate memory buffer. The experimental results show that this mechanism can reduce 20% ~70% stopping time of checkpointee to ensure the concurrency between doing checkpoint and execution of checkpointee. This mechanism is a quite good choice for hign real-time and interactive priority applications.