By Ben Bolles, VP of Product
Part 2 in a 3-part Series on NVMe technology and how Pivot3 uses NVMe to get the most out hyperconverged
Early adoption of HCI (hyperconverged infrastructure) by customers was limited to point-based, single applications, like VDI or light performance applications. As these early adopter customers began to realize the benefits of HCI (TCO reduction, simplified management, operational efficiencies, etc.), they were more confident in deploying HCI for broader applications and use-cases. Many are moving in the direction of deploying HCI for all of their mixed workloads, including performance-intensive applications. In order to meet these emerging customer demands, it is important that HCI solutions can deliver equivalent, or even better performance compared to the infrastructure the customer is currently using. Many of these customers have deployed virtualized servers for compute and all-flash/hybrid arrays for storage to meet their requirements. Bottom line: customers want to deploy HCI, but also need to ensure that it will perform for their applications.
This is why we at Pivot3 believe that the time is right for NVMe flash-enabled HCI to solve these challenges for customers. Not only can NVMe flash-enabled HCI solutions deliver better performance over traditional infrastructure approaches, the NVMe standard has emerged and is growing in popularity as more and more component and system vendors support NVMe-connected flash. Likewise, with an increasing emphasis on being “software-defined”, HCI, by definition, really comes down to solutions that are software-defined compute and storage combined. NVMe flash delivers the performance required, while being a standard interface that software solution vendors can build to.
In part one of this blog series – NVMe Technology Primer – I described the basics of NVMe technology and what is required to use NVMe (software and hardware). Building on that, for customers to take advantage of NVMe in a hyperconverged solution, three things need to be supported by the HCI vendor:
- Datapath architecture that can support NVMe flash
- Software interfaces that understand how to communicate via NVMe
- Qualified NVMe component and server platforms
Pivot3 has addressed all three of these key requirements in a way that gets the most out of NVMe technology for our HCI products. Ultimately, NVMe matters for HCI deployments for the following reasons:
1). NVMe delivers the performance required for mixed workload applications
In order to consolidate mixed application workloads onto an HCI solution, the solution must be able to deliver the IO and bandwidth performance required for those applications. NVMe flash enables an HCI solution to deliver the necessary performance so that customers don’t have to be concerned over the question of “will this new solution (HCI) perform as well, or better, than what I am currently running?” At Pivot3, we have measured our NVMe datapath flash compared to traditional SSD flash, and see up to 10X more IO performance from NVMe for the applications.
2). NVMe reduces Application Response Time
Many HCI v1.0 solutions (non-NVMe) really struggle to deliver performance for applications that require low response times, aka low latency. The industry standard by which many all-flash array vendors have succeeded is in delivering consistent sub-millisecond response times (latency) for applications. By bringing a NVMe flash datapath into HCI solutions, customers can get the low latency they require (sub-millisecond), which translates to reduced application response times. Pivot3 NVMe-enabled HCI products can consistently deliver sub-millisecond response times as measured at the application layer, which is what really matters to the end user.
3). NVMe flash datapath scales for performance better than SAS/SATA SSD flash datapath
Another property of the NVMe datapath is that it scales much farther and faster than older SAS/SATA connected technology. Part of this is due to the performance of PCIe connected flash vs. SAS/SATA connected flash. For example, an 8-lane PCIe NVMe flash card has ~63 Gb/sec available throughput compared to 12 Gb/sec for SAS SSD or 6Gb/sec for SATA SSD. This allows the NVMe flash datapath in an HCI solution to not only deliver more performance at lower latency, but to also scale that performance much farther before reaching a saturation point
Clearly, the benefits of NVMe are significant for our customers and will continue to be a key part of Pivot3 products and solutions as we move forward. We’ll discuss further details on how we have implemented our NVMe solutions here at Pivot3 in the next part of this blog series: Pivot3 use of NVMe – now and future.