``New'' Algorithm

1. The priority to be given to a page is not an intrinsic property but of the relation to which it belongs.
2. Each relation needs a working-set.
3. Buffer pool is divided and allocated on a per-relation basis.
4. Each active relation is assigned a resident set, which is initially empty.
5. Resident lists are arranged in a priority list, with a global free list on top.
6. The ordering is determined by a set of heuristics.
7. Search for a suitable buffer starts from the top of the priority list, and is added to the resident set of the relation.
8. MRU discipline followed within each resident set.
9. Each relation has at least one buffer which cannot be replaced.
10. Performance
    1. Results in paper showed throughput much better thab UNIX buffer manager.
    2. A trial implementation failedto improve the throughput against LRU implementation.
11. Drawbacks
    1. Use of MRU is justifiable in limited cases only.
    2. Rules for determining priority were based on intuition.
    3. Searching through a priority list can be expensive under high memory contentions.
    4. Extension to multi-user environment is difficult.