Lists Revisited: Cache Conscious STL Lists

Abstract

Abstract. We present three cache conscious implementations of STL standard compliant lists. Up to now, one could either find simple double linked list implementations that easily cope with standard strict requirements, or theoretical approaches that do not take into account any of these requirements in their design. In contrast, we have merged both approaches, paying special attention to iterators constraints. In this paper, we show the competitiveness of our implementations with an extensive experimental analysis. This shows, for instance, 5-10 times faster traversals and 3-5 times faster internal sort.

Locality-Conscious Lock-Free Linked Lists

“…Thus, a traversal of the list stays mostly within a chunk's boundary (a virtual page or a cache line), and therefore, the traversal enjoys a reduced number of page faults (or cache misses) compared to a traversal of randomly allocated nodes, each containing a single entry. Maintaining a linked list in chunks is often used in practice (e.g., [1,5]) but a lock-free implementation of a cache-conscious linked list has not been available heretofore. We believe that the building blocks of this list, i.e., the chunks and the freeze operation, can be used for building additional data structures, such as lock-free hash functions, and others.…”

“…For example, setting each chunk to be one virtual page, causes list traversals to form a page-oriented memory access pattern. This partition of the list into sub-lists, each residing on a small chunk of memory is often used in practice (e.g., [1,5]), but there is no lock-free implementation for such a list. Breaking the list into chunks can be trivial if there is no restriction on the chunk size.…”