Universal hashing (in a randomized algorithm or data structure) refers to selecting a hash function at random from a family of hash functions with a certain mathematical property (see definition below). This guarantees a low number of collisions in expectation, even if the data is chosen by an adversary. Many universal families are known (for hashing integers, vectors, strings), and their evaluation is often very efficient. Universal hashing has numerous uses in computer science, for example in implementations of hash tables, randomized algorithms, and cryptography.
Assume we want to map keys from some universe into bins (labelled ). The algorithm will have to handle some data set of keys, which is not known in advance. Usually, the goal of hashing is to obtain a low number of collisions (keys from that land in the same bin). A deterministic hash function cannot offer any guarantee in an adversarial setting if the size of is greater than , since the adversary may choose to be precisely the preimage of a bin. This means that all data keys land in the same bin, making hashing useless. Furthermore, a deterministic hash function does not allow for rehashing: sometimes the input data turns out to be bad for the hash function (e.g. there are too many collisions), so one would like to change the hash function.
The solution to these problems is to pick a function randomly from a family of hash functions. A family of functions is called a universal family if, .
In other words, any two keys of the universe collide with probability at most when the hash function is drawn randomly from . This is exactly the probability of collision we would expect if the hash function assigned truly random hash codes to every key. Sometimes, the definition is relaxed to allow collision probability . This concept was introduced by Carter and Wegman in 1977, and has found numerous applications in computer science (see, for example ). If we have an upper bound of on the collision probability, we say that we have -almost universality.
Many, but not all, universal families have the following stronger uniform difference property:
- , when is drawn randomly from the family , the difference is uniformly distributed in .
Note that the definition of universality is only concerned with whether , which counts collisions. The uniform difference property is stronger.
(Similarly, a universal family can be XOR universal if , the value is uniformly distributed in where is the bitwise exclusive or operation. This is only possible if is a power of two.)
An even stronger condition is pairwise independence: we have this property when we have the probability that will hash to any pair of hash values is as if they were perfectly random: . Pairwise independence is sometimes called strong universality.
Another property is uniformity. We say that a family is uniform if all hash values are equally likely: for any hash value . Universality does not imply uniformity. However, strong universality does imply uniformity.
Given a family with the uniform distance property, one can produce a pairwise independent or strongly universal hash family by adding a uniformly distributed random constant with values in to the hash functions. (Similarly, if is a power of two, we can achieve pairwise independence from an XOR universal hash family by doing an exclusive or with a uniformly distributed random constant.) Since a shift by a constant is sometimes irrelevant in applications (e.g. hash tables), a careful distinction between the uniform distance property and pairwise independent is sometimes not made.
For some applications (such as hash tables), it is important for the least significant bits of the hash values to be also universal. When a family is strongly universal, this is guaranteed: if is a strongly universal family with , then the family made of the functions for all is also strongly universal for . Unfortunately, the same is not true of (merely) universal families. For example, the family made of the identity function is clearly universal, but the family made of the function fails to be universal.
UMAC and Poly1305-AES and several other message authentication code algorithms are based on universal hashing. In such applications, the software chooses a new hash function for every message, based on a unique nonce for that message.
Several hash table implementations are based on universal hashing. In such applications, typically the software chooses a new hash function only after it notices that "too many" keys have collided; until then, the same hash function continues to be used over and over. (Some collision resolution schemes, such as dynamic perfect hashing, pick a new hash function every time there is a collision. Other collision resolution schemes, such as cuckoo hashing and 2-choice hashing, allow a number of collisions before picking a new hash function). A survey of fastest known universal and strongly universal hash functions for integers, vectors, and strings is found in.
For any fixed set of keys, using a universal family guarantees the following properties.
- For any fixed in , the expected number of keys in the bin is . When implementing hash tables by chaining, this number is proportional to the expected running time of an operation involving the key (for example a query, insertion or deletion).
- The expected number of pairs of keys in with that collide () is bounded above by , which is of order . When the number of bins, , is , the expected number of collisions is . When hashing into bins, there are no collisions at all with probability at least a half.
- The expected number of keys in bins with at least keys in them is bounded above by . Thus, if the capacity of each bin is capped to three times the average size (), the total number of keys in overflowing bins is at most . This only holds with a hash family whose collision probability is bounded above by . If a weaker definition is used, bounding it by , this result is no longer true. By avoiding modular arithmetic, this method is much easier to implement and also runs significantly faster in practice (usually by at least a factor of four). The scheme assumes the number of bins is a power of two, . Let be the number of bits in a machine word. Then the hash functions are parametrised over odd positive integers (that fit in a word of bits). To evaluate , multiply by modulo and then keep the high order bits as the hash code. In mathematical notation, this is
and it can be implemented in C-like programming languages by
(unsigned) (a*x) >> (w-M)
This scheme does not satisfy the uniform difference property and is only -almost-universal; for any , .
To understand the behavior of the hash function, notice that, if and have the same highest-order 'M' bits, then has either all 1's or all 0's as its highest order M bits (depending on whether or is larger). Assume that the least significant set bit of appears on position . Since is a random odd integer and odd integers have inverses in the ring , it follows that will be uniformly distributed among -bit integers with the least significant set bit on position . The probability that these bits are all 0's or all 1's is therefore at most . On the other hand, if , then higher-order M bits of contain both 0's and 1's, so it is certain that . Finally, if then bit of is 1 and if and only if bits are also 1, which happens with probability .
This analysis is tight, as can be shown with the example and . To obtain a truly 'universal' hash function, one can use the multiply-add-shift scheme
which can be implemented in C-like programming languages by
(unsigned) (a*x+b) >> (w-M)
where is a random odd positive integer with and is a random non-negative integer with . With these choices of and , for all . This differs slightly but importantly from the mistranslation in the English paper.
This section is concerned with hashing a fixed-length vector of machine words. Interpret the input as a vector of machine words (integers of bits each). If is a universal family with the uniform difference property, the following family (dating back to Carter and Wegman
In practice, if double-precision arithmetic is available, this is instantiated with the multiply-shift hash family of.
If double-precision operations are not available, one can interpret the input as a vector of half-words (-bit integers). The algorithm will then use multiplications, where was the number of half-words in the vector. Thus, the algorithm runs at a "rate" of one multiplication per word of input.
The same scheme can also be used for hashing integers, by interpreting their bits as vectors of bytes. In this variant, the vector technique is known as tabulation hashing and it provides a practical alternative to multiplication-based universal hashing schemes.
Strong universality at high speed is also possible. Initialize the hash function with a vector of random integers on bits. Compute
The result is strongly universal on bits. Experimentally, it was found to run at 0.2 CPU cycle per byte on recent Intel processors for .
This refers to hashing a variable-sized vector of machine words. If the length of the string can be bounded by a small number, it is best to use the vector solution from above (conceptually padding the vector with zeros up to the upper bound). The space required is the maximal length of the string, but the time to evaluate is just the length of . As long as zeroes are forbidden in the string, the zero-padding can be ignored when evaluating the hash function without affecting universality. treats the string as the coefficients of a polynomial modulo a large prime. If , let be a prime and define:
- , where is uniformly random and is chosen randomly from a universal family mapping integer domain .
Using properties of modular arithmetic, above can be computed without producing large numbers for large strings as follows:
uint hash(String x, int a, int p) uint h = INITIAL_VALUE for (uint i=0 ; i < x.length ; ++i) h = ((h*a) + x[i]) mod p return h
This Rabin-Karp rolling hash is based on a linear congruential generator. Above algorithm is also known as Multiplicative hash function. In practice, the mod operator and the parameter p can be avoided altogether by simply allowing integer to overflow because it is equivalent to mod (Max-Int-Value + 1) in many programming languages. Below table shows values chosen to initialize h and a for some of the popular implementations.
|Bernstein's hash function djb2||5381||33|
|Kernighan and Ritchie's hash function||0||31|
Consider two strings and let be length of the longer one; for the analysis, the shorter string is conceptually padded with zeros up to length . A collision before applying implies that is a root of the polynomial with coefficients . This polynomial has at most roots modulo , so the collision probability is at most . The probability of collision through the random brings the total collision probability to . Thus, if the prime is sufficiently large compared to the length of strings hashed, the family is very close to universal (in statistical distance).
Avoiding modular arithmetic
To mitigate the computational penalty of modular arithmetic, three tricks are used in practice:
- One chooses the prime to be close to a power of two, such as a Mersenne prime. This allows arithmetic modulo to be implemented without division (using faster operations like addition and shifts). For instance, on modern architectures one can work with , while 's are 32-bit values.
- One can apply vector hashing to blocks. For instance, one applies vector hashing to each 16-word block of the string, and applies string hashing to the results. Since the slower string hashing is applied on a substantially smaller vector, this will essentially be as fast as vector hashing.
- One chooses a power-of-two as the divisor, allowing arithmetic modulo to be implemented without division (using faster operations of bit masking). The NH hash-function family takes this approach.