Paper Title: Bare Metal Performance for Software Defined Storage with NVMe

Paper Abstract: NVMe presents many challenges to developers of software defined storage, particularly in a hyperconverged architecture. Although reducing context switches, thread synchronization, and memory copying in the storage stack can provide much lower latency and greater overall throughput in synthetic benchmarks, efficient management of CPU resources is equally important when maximizing throughput for real applications. In practice, the software storage stack must be adapted to NVMe in such a way as to take full advantage of the hardware capabilities while preserving CPU cycles for primary workloads.

Paper Author: Phil White, CTO, Scale Computing

Author Bio: Phil White is CTO of Scale Computing, where he focuses on designing and implementing complex systems such as SCRIBE, a distributed software-defined block storage engine. He is an expert in the design, development, and maintenance of multithreaded userspace daemons in Linux. He has 16 years of experience in the software industry during which he has written everything from kernel code to user interface code plus all that lies in-between. He earned a BS in computer science from the Rose-Hulman Institute of Technology (Indiana). He is a member of the Forbes Technology Council and has spoken at several conferences, including OpenStack Summit.