Fault Interpretation: Fine-Grain Monitoring of Page Accesses

This paper presents a technique for obtaining ne-grain information about page accesses from standard virtual memory hardware and Unix operating system software. This can be used to monitor all user-mode accesses to speciied regions of the address space of a process. Application code can intervene before and/or after an access occurs, permitting a wide variety of semantics to be associated with memory pages. The technique facilitates implementing complex replication or consistency protocols on transparent distributed shared memory and persistent memory. The technique can also improve the eeciency of certain generational and incremental garbage collection algorithms. This paper presents our implementation and suggest several others. EEciency measurements show faults to be about three orders of magnitude more expensive than normal memory accesses, but two orders of magnitude less expensive than page faults. Information about how to obtain the code via anonymous ftp appears at the end of the paper.