Concurrent Checkpointing and Recovery in Distributed Systems
Pei-Jyun Leu, Bharat Bhargava · Purdue e-Pubs (Purdue University System) · 1987
This paper studys concurrency issues in disUibuled checkpointing and rollback recovery. It transforms the concurrent checkpointing and recovery problem to a transaction processing problem. A new transaction model, which consists of four types of atomic operations and five types of conflicts, is used for disuibuted checkpointing and recovery. Each transaction is executed by multiple processes in the system. We have shown that the consistency of recovery lines and rollback lines established by checkpoint transactions and rollback transactions can be achieved by enforcing serializability on the corresponding lransaclions. An algorithm is designed to expand and execute checkpoint transactions or rollback transactions concurrently. The algorithm supports efficient recovery, reduces the response time of checkpoint transactions and rollback transactions, and allows nonnal messages to be transmitted in any order. We have implemented the algoritlun for perfonnance evaluation. The analysis shows thaI concurrent execution reduces the response time of checkpoinL transactions and rollback transactions. The CPU cost is in a linear order of the tOlal number of synchronization messages used. For a checkpoim/rollback transaction with eight participating processes, the CPU cost is significantly smaller than the single checkpoint/rollback cost when the processes are bigger than 12K bytes. This wode was supported in part by UNTSYS. and NASA, and a David Ross fellowship. U An earlier version of this paper appears in Prot', IEEE 4t1l Con{. Da/a Engineering, Los Angeles. CA, Feb. 1988.