Paper Title: Using Multi-Drive Fusion to Scale NVMe Performance

Paper Abstract: NVMe SSDs provide high I/O performance and greatly alleviate the I/O bottleneck in computer systems. However, it is still difficult to scale I/O performance from multiple NVMe drives while maintaining redundancy among them. A new solution based on what we call multi-drive fusion can help solve the problem. It involves having multiple NVMe drives, each with a properly enhanced controller, collaborate without requiring many system CPU cycles or much memory space. Multi-drive fusion balances the I/O workload automatically among the drives,and also levels wearing of the drives by distributing writes properly. It generates data parities as needed by using the accelerators built into the drive controllers. With multi-drive fusion, a system can aggregate the performance of all installed NVMe drives efficiently and run at a configurable redundancy without straining host resources.

Paper Author: Jinling Chen, Field Engineering Manager, UNIC

Author Bio: Jinling Chen is Field Engineering Manager at UNIC² Memory Technology. He is responsible for developing solutions and providing technical support for enterprise NVMe SSD products. Before joining UNIC², he worked on microprocessor design, embedded system software design, and NOR/NAND flash based storage solutions. He was a key design team member for the first RISC processor in China, and proposed the concept and led the design of a hybrid server memory subsystem based on high-density NOR flash. He earned an MS in Microelectronics from Peking University.