fast-skiplist | A fast , threadsafe skip list in Go

fast-skiplist is a Go library. fast-skiplist has no bugs, it has no vulnerabilities, it has a Permissive License and it has low support. You can download it from GitHub.

"Perfection is achieved not when there is nothing more to add, but rather when there is nothing more to take away" — Antoine de Saint-Exupery. If fast-skiplist is faster than other packages with the same features, it's because it does less wherever possible. It locks less, it blocks less, and it traverses less data. Even with these tricks up its sleeve, it has fewer lines of code than most implementations. When inserting, it calculates "height" directly instead of consecutive "coin tosses" to add levels. Additionally, it uses a local PRNG source that isn't blocked globally for improved concurrent insert performance. The probability of adding new nodes to each level of the structure (it's height) is determined by successively "rolling the dice" at each level until it doesn't meet a fixed value P. The default P values for skip lists in the wild range from 0.25 to 0.5. In this implementation, the default is 1/e, which is optimal for a general-purpose skip list. To find the derivation of this number, see Analysis of an optimized search algorithm for skip lists Kirschenhofer et al (1995). Almost all other implementations are using common functions in math/rand, which will block because querying the PRNG to determine the height of new nodes waits then acquires a lock via the system-wide random number generator. We get around this by assigning a new rand source to each skip list instantiated, so each skip list can only ever block itself. This significantly speeds up insert times when you are managing multiple lists with high concurrency. Additionally, this implementation always requests just one number from the PRNG. A pre-computed probability table is used to look up what the height of the new node will be. This is faster and offers a fixed calculation time compared to successive "dice rolls" for each level. The table is computed for each level L using the default P value of 1/e: math.Pow(1.0/math.E, L-1). It is consulted during inserts by querying for a random number in range [0.0,1.0) and finding the highest level in the table where the random number is less than or equal to the computed number.