| 摘要 |
In a fault-tolerant computer system, a primary replica supervisor is interposed between an operating system and a primary replica of an application program being executed by a primary processor. An object-code editor locates calls to the operating system and loops in the application program and inserts instruction sequences that enable the replica supervisor to intercept the calls to the operating system, results returned by the operating system as a result of the calls and asynchronous events delivered by the operating system to the replica. A backup replica supervisor is similarly interposed between an operating system and a backup replica of the application program being executed by a backup processor. The primary replica interacts with an environment. The replica supervisors ensure that the backup replica undergoes state transformations, as a result of the calls to the operating system and asynchronous events, that are equivalent to state transformations that the primary replica undergoes as a result of corresponding calls and asynchronous events. Thus, after a failure in the primary processor, the backup replica can interact with the environment in a manner consistent with interactions between the primary replica and the environment prior to the failure. |
