Towards Merkle Trees for High-Performance Data Systems

Abstract

Merkle Trees (and its variants) are widely used for building secure outsourced data systems. The adoption of Merkle Trees for high-performance data systems, however, uncovered major performance challenges. First and unlike classical data structures, Merkle Trees involve expensive cryptographic operations and are thus CPU-bound. Second, they are not well suited for modern multi-core CPUs because they introduce a single point of contention making Merkle Trees hard to parallelize. While recent work aimed at replacing Merkle Trees to circumvent their performance problem, we suggest new techniques to speed-up this ubiquitous data structure and achieve high-performance. In this paper, we present initial results showing that in contrast to common wisdom it is indeed possible to build high-performance Merkle Trees with orders of magnitude performance improvements.