Lightweight Write Detection and Checkpointing for Fine-Grained Persistence
Antony L. Hosking, J. Eliot B. Moss · Purdue e-Pubs (Purdue University System) · 1995
Many systems must dynamically track writes to cached data, for the purpose of reconciling those updates with respecL to the permanent or global staLe of Lhe daLa.For eX<lmple, distribmed systems employ coherency protocols to enSUre a consi5tenl vie ..... ofshared data.Similarly, database 5ystems log updates both for concurrency control and La ensure the resilience of Lhose update$ in the face of sysLem failures.Here, we measure and compare the absolute perforffi<lnCe of several alternative mechanism5 for the light\\'eight detection of wriles to cached data in a persistent syslem, and the relath'e overhead to log those writes to stable storage in the form of a chct;kpoint.A chccltpoint defines a consistent slate to which the system will be restored in the evcnt of any subsequent failUre.The efficient detection and logging of updatcs is critical to the performance of persislent systems that embody a fine--g.-aineddata model, since per-object overheads are lypic<l.ll.v very low.Our results reve.u a wide range of performance for lhe altern