Workload Analysis of Key-Value Stores on Non-Volatile Media

Author(s)/Presenter(s):
Library Content Type:
Publish Date: 
Wednesday, September 13, 2017
Event Name: 
Focus Areas:
Abstract: 

Key-value (KV) stores are an important category of NoSQL databases that has gained popularity for high-performance Cloud workloads such as social networks, gaming and data analytics. Popular design choices for persistent KV stores are LSM and B-Tree – we put these approaches head-to-head in terms of performance – CPU load profile, throughput, latency and IO pattern at the device level, and space-amplification – from leveled/tiered compaction in case of LSM and from fragmentation with B-Tree-based designs. We’ll discuss trade-offs of the LSM and B-Tree approaches for NAND media and the performance improvements that 3D Xpoint SSDs bring.

Watch video: