Part II: HASHING FOR STORAGE: DATA MANAGEMENT
CHAPTER 10
Record Chaining with Hash Tables
10.1 SEPARATE CHAINING OF RECORDS
The cost of SEARCH, as measured by the number of comparisons, when N keys are stored as in Example 7.1, in a single chain is O(N). A significant improvement might be expected by combining record chaining with hashing, which results in chains rooted at hash-table cells. When a single chain of N records is replaced by r chains 0, 1, ··· , r−1 of lengths L0, L1, ··· , Lr−1 with N = L0 + L1 + ··· + Lr−1, the number of comparisons required in SEARCH or INSERT is reduced.
Knuth [Knuth 1973, pp. 513–14] referred to the merge of hashing and chaining as separate chaining; the idea was invented by H.P. Luhn in 1953 (see Chapter 8) and described in memoranda dated 1/3/53 and 2/20/53 to Mr. J. C. McPherson, who then an IBM Vice-President, Luhn wrote1
When in loading the system, an address is found to be occupied by a previous entry, the address number of the next auxilliary address is recorded behind the original address and the new item is recorded at this auxilliary address. ...