Paper Title: Using Storage Accelerators to Provide Predictable NVMe SSD Performance

Paper Abstract: Most applications today run on virtual machines (VMs) and cloud providers become more dependent on the predictable performance of their VM infrastructure. However, such machines assume that underlying resources such as SSDs are predictable in performance (otherwise, overall results could vary considerably depending on which resources were assigned to a task and how they behaved). In practice, SSD performance (particularly latency) in multi-tenant environment is not totally predictable due to the way in which the underlying flash is managed, the technology that is used, and the OS overhead. Storage accelerators can alleviate the resulting challenges by moving overhead tasks into hardware and simplifying the VM OS (hypervisor) management, hence providing better SSD quality of service (QoS). Cloud providers will be able to better control their infrastructure bandwidth and minimize unpredictable storage performance.

Paper Author: Shahar Noy, Sr Director Product Marketing, Marvell

Author Bio: Shahar Noy is Sr Director of Product Marketing for Data Center Storage Solutions at Marvell. He leads a team focused on the definition, planning, development, and marketing of the company’s cutting-edge products and solutions. He emphasizes the enhancement of storage performance and efficiency for both cloud and enterprise applications. He is also involved in working with emerging technologies and technology partners. Before joining Marvell, Shahar was Sr Director or Strategic Marketing and Business Development at Micron, where he worked on new consumer and enterprise SSDs products and led the effort toward the first removable NVMe solution. He also worked at SanDisk, where he helped win early TLC NSAND designs. He earned an MBA at the University of Southern California’s Marshall School of Business and a BSEE from the Technion-Israel Institute of Technology. He has over 20 years experience in the technology industry.