Abstract
When providing data management for nontraditional data, database systems encounter storage reclamation problems similar to those encountered by virtual memory managers. The paging behavior of existing automatic storage reclamation schemes as applied to objects stored in a database management system is one indicator of the performance cost of various features of storage reclamation algorithms. The results of modeling the paging behavior suggest that Mark and Sweep causes many more input/output operations than Copy-Compact. A contributing factor to the expense of Mark and Sweep is that it does not recluster memory as does Copy-Compact. If memory is not reclustered, the average cost of accessing data can go up tremendously. Other algorithms that do not recluster memory also suffer performance problems, namely all reference counting schemes. The main advantage of a reference count scheme is that it does not force a running program to pause for a long period of time while reclamation takes place, it amortizes the cost of reclamation across all accesses. The reclustering of Copy-Compact and the cost amortization of Reference Count are combined to great advantage in Baker's algorithm. This algorithm proves to be the least prohibitive for operating on disk-based data.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have