MultiPath MultiGet: An Optimized Multiget Method Leveraging SSD Internal Parallelism

Abstract

Accessing data with low latency is a major issue for big data platforms. Some platforms use a Multiget method when accessing data. The Multiget method can be used efficiently when a server should seek data with multiple keys or many single get requests are waiting in a command queue. The method is implemented to access multiple blocks of data using multiple keys. In this research, we developed MPMG, a novel multipath Multiget method for a solid state drive (SSD)-based key-value storage. In MPMG, we use the open source key-value storage RocksDB, which was developed by Facebook and is one of the most representative key-value NoSQL stores. Current RocksDB system cannot fully utilize the internal parallelism of SSD. The MPMG method was designed to leverage ssd internal parallelism and access data in SSD with parallel approach. The ability to support the parallel access inside SSD is efficient for performance. With multipath approach, the SSD can fully exploit its bandwidth, compared to the original approach. In an experiment, this method showed improvements in speed and throughput. The elapsed time for processing decreased up to 80.