Algorithms for memory hierarchies: advanced lectures
Algorithms for memory hierarchies: advanced lectures
| Hi-index | 0.00 |
Systems that buffer data in buffer pools allocated in virtual memory can experience the double paging anomaly, which may result in significant performance degradation. While static memory allocation techniques have been used to successfully avoid these anomalies, they are not adequate when either dynamic memory allocation is used, or large buffer pools are required. In this paper, we consider ways of avoiding the double paging anomaly when memory is dynamically allocated. Our approach is based on using an appropriately sized buffer pool. We argue that only two schemes are appropriate in such an environment: one involves using an arbitrary pool size, while the other requires the pool size to be as large as the database. The former scheme involves using an extended virtual memory interface to eliminate replacement faults and places extremely strict requirements on the nature of the replacement policies used at both levels. The latter scheme, however, is more general, and our results indicate that setting the pool size to that of the database size represents the best way to avoid or reduce the anomaly, when dynamic allocation is used. This scheme is also attractive in light of the availability of large virtual memory address spaces. We describe the class of systems for which these schemes are most appropriate for and suggest ways for realizing them. We also show how such systems can be used to model and reinforce the case for systems that map files into virtual memory.