Combinatorial input recognition algorithm Internet applications to HTTP web servers

Abstract

We introduce an algorithm for the classical recognition problem of identifying elements s within a class of strings S = {s1, s2, ..., sn}. Here it is assumed that struct s = { char *a; int z; }; and that for two strings s1, s2 if s1.z == s2.z then there is J = {j1, j2, ..., jl} with l(z) < z so that s1[J] == s2[J] ⇒ s1 == s2. The goal is to minimize Σ = n k i l si z 1 ( . ) and thus distinguish elements with the smallest number of comparisons. In general, this simply means that in a language system we face a nonoptimal set of keywords (n < sizeof(*s.a)z) and so we seek to reconstruct the language with a smaller set of k recognition vertices so that k ≈ logsizeof(*s.a) n. The algorithm is applied to increasing HTTP transmission and request processing speed. The result is an increase in web server performance from 20,000 to 35,000 or more hits per second. The input recognition algorithm produces the greatest improvement when interpreting HTTP request headers which are large in proportion to the response. This is the case when collecting data for statistical analysis of Internet traffic.