Parsing gigabytes of JSON per second with parallel bit streams

Studies have shown that it is possible to boost the efficiency of text processing by carefully eliminating branches as well as reducing branch mispredictions and cache misses, which can be achieved with a few techniques, such as the use of Boolean algebra to reduce pointer-chasing in data structures and to abstract branching. With current advances in technology, vector extensions (SIMD) have been added to commodity processors and have allowed the creation of new algorithms that are able to accomplish the non-trivial task of parallelly processing streams in Gigabytes per second. The Parabix framework exploits the concept of parallel bit streams to take even more advantage of SIMD instructions by transposing and processing streams in batches. This study focuses on using Parabix to boost the efficiency of JSON parsing for Big Data.