Accelerating Data Ingress for Range-Scan Optimized HBase Instances

HBase was designed to get high performance for read-intensive workloads, making it very attractive for use cases involving retrieval of contiguous sets of rows. HBase instances set up for fast reads often force users to compromise on write performance, which in turn can lead to temporary degradation of read capabilities. In this paper we study existing write penalty mitigation options and introduce our software solution, which not only accelerates bulk data ingress but also retains egress performance without any post-processing or alterations of the soft-schema. Our experiments show that the proposed software solution yields a 47 % reduction of ingestion duration.