LINEAR HASHING

1
LINEAR HASHING

• Prepared by Vijay T. Raisinghani
• KRESIT, IIT, Bombay

2
Introduction

• Tree works reasonably well in case of dynamic
  files though requiring several accesses
• Dynamic and Extendible hashing require at least
  two accesses since the data structures for the
  dynamically created hashing functions used must
  be on the disk
• Linear Hashing requires a few bytes of main
  storage for a file of any size

3
Principles

• Overflow records lead to rapid deterioration in
  performance as buckets fill up
• Modification to the hashing function and file is
  required called a split
• In Linear hashing this split results in linear
  increase in the address space
• The basic split functions are
• hiC--gt0,1,.., 2iN-1 1 and for any c hi(c)
  hi-1(c) 2.1
• or hi(c) hi-1(c) 2i-1N 2.2

4
Principles
5
Principles

• The split new bucket, is the primary bucket and
  the records are accessible in one access
• The splits are done in a pre-defined order and
  not every time a collision occurs
• A pointer indicates location of the split
  (initially 0). If the collision position is after
  this pointer, classical collision resolution (eg
  chaining) is used
• This bucket will have a delayed split

6
Performance Analysis

• The split (or grouping) may be controlled by way
  of maintaining the load factor at a threshold
• High and constant load factor implies minimal
  storage and faster access
• The algorithm for insert/delete/concurrency
  control is simpler, as lock on the pointer and
  key only is required ( not on a sub-tree)

7
Applications

• Useful where the file may grow or shrink heavily
• Eg Compilers / text processing systems
• Avoids difficult estimation of DBMS files where
  deleted records may not be physically deleted
• Good tool for management of workspaces of queries
  and even virtual memory
• Since it works with very small cores, LH renders
  dynamic files usable on micro-computers