FMS: By GigaIO, FabreX Implementation of NVMe-oF Architecture

GigaIO Networks, Inc. introduced the FabreX implementation of NVMe-oF architecture, streamlining NVMe network communication and large-scale storage sharing with low latency and high-bandwidth features.

FabreX

NVMe is an open interface for accessing non-volatile storage media attached via PCIe. This scalable interface is designed for use in PCIe-based systems, facilitating expedited access to direct-attached SSDs.

Similarly, NVMe-oF is used for network communication between multiple hosts, extending NVMe architecture access fabric technologies to enable expedited access between servers and NAS.

“FabreX is the next generation PCIe-based network, and with our new NVMe-oF support, we bring the industry’s lowest latency and highest throughput to data centers for quick and efficient data transfers,” says Alan Benjamin, CEO, GigaIO. “With FabreX and NVMe-oF, storage located in a server, directly attached to a server and attached across a network delivers identical performance, allowing customers to choose their preferred storage architectures and experience the same QoS to meet desired service level objectives.“

The company’s implementation of FabreX with NVMe-oF allows devices and data to remain on native PCIe networks and NVMe protocols, respectively, to sustain optimum performance. In contrast, other NVMe-oF options convert device connections between PCIe and Ethernet or FC and IB fabrics, and data paths between NVMe and RoCE or iWARP protocol. These conversions create unavoidable inefficiencies that lower potential performance. For example, while NVMe-oF storage latency is customarily measured in microseconds, when used with Ethernet or FC fabrics, it is best measured in nanoseconds, specifically when using NVMe-oF with FabreX native PCIe networks.

Click to enlarge

The benefits of a native NVMe-oF implementation with FabreX include low latency and high throughput, extended direct data exchange and reduced CPU resource utilization. Additionally, FabreX with NVMe-oF allows direct data placement between NVMe drives and remote host systems memory, eliminating all buffers and copies. This enables rapid communication between local host memory and remote NVMe subsystems, outperforming all other fabrics.

“With the explosion of data that customers are now gathering, and the subsequent need to analyze that data and turn it into meaningful information to drive their business, companies tell us they have a burning need for higher performance and better utilization of their compute, network and storage investments,” continues Benjamin. “Often, the biggest complaint is the legacy networks that tie their systems together. The truth is the legacy networks were never designed for the rack-scale systems people now need and use, and while they are trying their best to respond, a fresh approach to creating these systems is needed.“

FabreX is a network architecture for compute and storage infrastructure. The platform delivers performance with low latency and flexibility, allowing data centers to boost their throughput while driving down both their capital and operating costs for a TCO.

Availability
FabreX for NVMe-oF will become a standard part of the company’s FX/OS software package and will be released for availability by the end of September 2019.

Resource:
New Frontiers in NVMe-oF

About GigaIO
GigaIO Networks, Inc. is creating the next gen network providing hyper-performance and flexibility. The result is optimized rack and cluster system performance and a reduced TCO. With the FabreX architecture, data centers can scale up or out the performance of their systems, preserving their existing investment while adding the newest and latest capabilities.