MIGSOCK vs . Zap ∗

This paper demonstrates the performance gain of MIGSOCK over Zap, an user-level approach to network process checkpointing and restarting. We consider two scenarios where process checkpointing and restarting could take place. In the first scenario, where processes are aggregated in one PrOcess Domain (pod), it takes Netfilter three times as much time as MIGSOCK to handle translation and forwarding of 10,000 packets. In the multiple-pod scenario, modeling multiple-service support, assuming the time to deliver a pod is constant, it takes about 10 pods for Netfilter to break even with MIGSOCK. In other words, MIGSOCK only suffers a performance hit in checkpointing 10 pods and beyond. We argue that this scenario can be avoided by load balancing services with a threshold of 10 per server. Based on empirical analysis, we conclude that MIGSOCK is an ideal solution to network process checkpointing in comparison to Zap.