Skip to content

Weaving Relations for Cache Performance (Ailamaki, VLDB 2001) #13

@benclmnt

Description

@benclmnt

This paper proposes a new page layout which improves cache efficiency.

image

The classic "slotted page" layout is shown above. It is great for accessing data record-by-record, but it exhibits poor cache usage during predicate evaluation because it brings other unused data into the cache line.

image

Compare this to PAX, where subsequent data are passively "prefetched" as it is brought together into the cache. This lead to better cache efficiency. However, PAX needs to reconstruct tuples, although this cost is minimal (no changes in IO).

PAX contains the same set of tuples as classic "slotted page", so it exhibits the same IO patterns. It only reorders the inner page layout.

Question:

  • It makes sense for this layout to be used in OLAP workloads (columnar storage). But why is this page layout not used by all OLTP databases as well?

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions