Separately manage each type in its domain of associated buffers.
Inside each domain buffers are managed by the LRU principle.
Type assignment scheme :
Assign one domain to each non-leaf level of the B-Tree structure.
Another to the leaf level together with the data.
Performance : 8-10 % improvement in throughput over the conventional LRU.
Drawbacks :
Domain is static : Importance of a page may vary in queries.
No differentiation in the relative importance between different
types of pages.
Partitioning buffers into domains, rather than queries, makes the
buffers susceptible to interference among competing users.
No built-in facilities for load control.
Extensions :
Group LRU : Assigns a fixed priority ranking among different
domains. Search for a free buffer always starts from the domain with
the lowest priority.
Working-Set : The size of a domain is dynamically varied using a
working-set like partitioning scheme. Pages in domain i which had
been referenced in the last references were exempt from the
replacement considerations.