MC 302 – DBMS: Hashing

Goonjan Jain
Department of Applied Mathematics
Delhi Technological University
Outline
• hashing
• extendible hashing
• linear hashing
• Hashing vs B-trees
Hashing
• Primary file organization
• Very fast access to records on search condition
• Search condition must be on a single field – hash field or hash key
• Hash function/ randomizing function – applied on hash field, yields
address of disk block
Static Hashing
• Fixed number of buckets
• Drawback for dynamic files
• Types:
• Internal Hashing
• External Hashing
Internal Hashing
• For internal files
• implement a hash table using array of records
• M slots addressed as 0 to M-1
• Choose a hash function, transform hash field into an integer from 0 to
M-1
• Most common hash function h(k) = k mod M
• Problems:
• No guarantee that different values will hash to different addresses
Collision
• Hash field value hashed to an address already occupied
• Collision resolution: find another position
• Collision Resolution Techniques:
• Open addressing:
• Starting from the hashed address, check subsequent positions till an unused position is
found
• Chaining:
• Place new value in an unused overflow location
• Set a pointer from the address to the overflow location

• Multiple Hashing:
• Apply a second hash function
External Hashing
• Target address space is made of buckets,
• each bucket can hold multiple values
• Bucket is
• 1 disk block, or
• Cluster of contiguous blocks
• Hash function maps key to a relative bucket number
• Collision problem is less severe
• If a bucket if full, a variation of chaining can be used-
• Bucket points to record pointers – block address and record position
Problem with (static) hashing
• Overflow
• Underflow
• Sol: Dynamic Hashing
Dynamic Hashing
• idea: shrink / expand hash table on demand..
• ..dynamic hashing
• Details: how to grow gracefully, on overflow?
• Many solutions – One of them: ‘extendible hashing’
Extendible Hashing
• keep a directory, with pointers to hash-buckets
• Uses-
• An array of 2d buckets. d is the global depth
• Directory
• First d bits of hash value is used as index for directory entry
• Entry in director determines the bucket address
• Each bucket stores
• Local depth d’ – number of bits on which bucket contents are based
• Q: how to divide contents of bucket in two?
• A: hash each key into a very long bit string; keep only as many bits as needed
Extendible Hashing
Extendible Hashing
Extendible Hashing
Extendible Hashing
Extendible Hashing – Directory Doubling
• bucket overflows, 2 cases –
• If local depth, d’ = global depth, d
• Double the directory
• If d’< d
• Split the bucket, no need of directory doubling
Extendible Hashing
• Advantages:
• Performance of file does not degrade with increase in size
• Space overhead for directory is negligible
• Splitting causes minor reorganization
• Disadvantages:
• Directory must be searched before bucket. 2 block access instead of 1
Linear hashing
• Motivation: extendible hashing needs directory which doubles
• Q: can we do something simpler, with smoother growth?
• A: split buckets from left to right, regardless of which one overflowed
Initially: h(x) = x mod N (N=3 here)
Assume capacity: 2 records / bucket
Use two hash functions:
h0(x) = x mod N (N=3 here) - for unsplit buckets
h1(x) = x mod (2*N) (N=3 here) - for the splitted ones
Linear Hashing Example
• In the following M=3 (initial # of buckets)
• Each bucket has 2 keys. One extra key for overflow.
• s is a pointer, pointing to the split location. This is the place where
next split should take place.
• Insert Order: 1,7,3,8,12,4,11,2,10,13
• After insertion till 12:
Linear Hashing example
• When 4 inserted overflow occurred. So we split the bucket (no matter it is full or partially empty).
And increment pointer.

• split bucket 0 and rehashed all keys.

• Insertion of 10:
• (10 mod 3 = 1) and bucket 1 < s, we need to hash 10 again using h 1(10) = 10 mod 6 = 4th bucket
• For 13
• same bucket
• Overflow
• split 2nd bucket.
Linear Hashing Example
• Final Hash Table:

• s is moved to the top again as one cycle is completed and level is
incremented.
Linear hashing - Searching
• Algo to find key ‘k’:
compute b= h0(k) // original slot
if b < s // has already split
compute b= h1(k)
search bucket b
Linear hashing - Deletion
• Inverse of insertion
• If underflow, contract
• If the last bucket is empty,
• remove it and
• Decremented s.
• If s is 0 and the last bucket becomes empty,
• s is made to point to bucket (n/2)-1, where n is the current number of
buckets,
• Level is decremented, and
• the empty bucket is removed.
B+ Trees vs Hashing
B+ Trees Hashing
• Speed on Search • Speed
• Exact match queries, worst • On exact match queries, on
case the average
• Range queries
• Nearest-neighbor queries
• Speed on insertion + deletion
• Smooth growing and shrinking
(no-reorg)