B-link indexes are a variant of B- and B+- tree indexes in which each index page, non-leaf and leaf, contains the page number of (or a link to) its right sibling. Further, index pages need not appear immediately in a parent page. The primary advantage of B-link indexes is improved concurrency.
To improve fanout, SQL Anywhere stores a compressed form of each indexed value in which the prefix shared with the immediately preceding value is not stored. To reduce the CPU time when searching within a page, a small look-aside map of complete index keys (subject to data length restrictions) is also stored. In particular, SQL Anywhere indexes efficiently handle index values that are identical (or nearly so), so common prefixes within the indexed values have negligible impact on storage requirements and performance.