OCP Summit: Cavium Collaborating With Microsemi and Marvell for Compute, Networking and I/O for JBOF/FBOF Solutions

Cavium, Inc. announced that it is collaborating with Microsemi Corporation, and Marvell Technology Group Ltd. to demonstrate a reference architecture for accelerating and scaling out NVMe over Fabrics (NVMe-oF).

The architecture was demonstrated live on ARM64 and x86 architectures at the OCP Summit 2018 in San Jose, CA.

Integrating a range of Cavium technologies (ThunderX2 processors, FastLinQ 100GbE NICs, XPliant programmable switches) with Microsemi’s Switchtec PCIe switches and Flashtec NVRAM cards and Marvell’s NVMe SSD controllers enables multiple differentiated and validated end-to-end NVMe solutions, including NVMe-based all-flash arrays, just a bunch of flash and fabric-attached bunch of flash.

In addition to a hardware solution, the collaboration enables software integrations of Microsemi’s open source PCIe peer-to-peer (P2P) technology and Cavium FastLinQ concurrent RoCE and iWARP for Storage Performance Dataplane Kit (SPDK). This solution enables efficiencies and performance by offloading the compute resources from the data path leveraging NVMe 1.2 Controller Memory Buffer (CMB) technology.

At the OCP Summit 2018, Cavium, Microsemi and Marvell demonstrated concurrent 100Gb/s NVMe-oF RoCE and iWARP connectivity from FastLinQ 100GbE OCP NICs on x86 servers to a Celestica Nebula JBOF (with Marvell NVMe SSD controllers) and Facebook Lightning JBOF driven by Cavium’s ThunderX2 and XPliant platforms. The demonstration showcases how storage IHVs can leverage PCIe P2P and SPDK technology to completely offload the server CPU from the datapath and reduce latency.

“NVMe technology is rapidly changing the way cloud and enterprise data centers connect to shared storage, but building NVMe-oF solutions require industrywide technology integrations,” said Christopher Moezzi, VP, marketing, Ethernet server adapter group, Cavium. “Cavium FastLinQ Universal RDMA technology combined with the broader Cavium and Microsemi portfolio delivers a high performance and validated reference platform for JBOF/FBOF implementations, reducing risk and time to market for joint storage customers.“

“We are excited to collaborate with Cavium as part of Microsemi’s Accelerate Ecosystem. Microsemi’s leadership in NVMe connectivity and peer-to-peer technology reduces time to market and time to revenue for our joint customers in creating NVMe-oF solutions,” said Andrew Dieckmann VP, marketing, Microsemi. “The scale-out capabilities of Microsemi’s Switchtec PCIe switches and P2P data plane offload combined with Cavium’s unique concurrent RoCE and iWARP NICs provides compelling flexibility, scale and performance for enterprise data center and cloud applications.“

“Marvell’s NVMe cloud and enterprise SSD controller data center customers are seeking to scale and disaggregate their storage solutions without compromising latency, performance or reliability,” said Nigel Alvares, VP, SSD and data center storage solutions, Marvell. “Cavium’s unique Universal RDMA capabilities built into the FastLinQ 41000 Series Ethernet NICs, combined with its integration across the broad Cavium portfolio, are well-positioned to meet the requirements of next-generation data centers while offering best-in-class TCO.“

Cavium FastLinQ NVMe over Ethernet Universal RDMA solutions for NVMe-oF

FastLinQ adapter

Driven by the performance demands of NVMe, performance, low latency networking is a fundamental requirement for a fabric to scale out. Ethernet-based RDMA fabrics with their low latency and offload capabilities have become a choice for NVMe-oF. FastLinQ 10/25/40/50/100GbE Ethernet adapters support Universal RDMA (RoCE, RoCEv2 and iWARP) and deliver a choice to customers for scaling out NVMe over a general-purpose Ethernet fabric. In addition, LiquidIO line of SmartNIC solutions are key enablers for delivering JBOF/FBOF management and AFA storage services like full IPSEC offloads, compression and deduplication.

LiquidIO-II adapters.

Cavium ThunderX2 servers solutions for NVMe and NVMe-oF
The ThunderX2 product family is the company’s second-generation, 64-bit ARMv8-A server processor SoCs for the data center, cloud, storage and high-performance computing applications. The family integrates performance custom cores along with outstanding I/O connectivity, memory bandwidth and capacity. Delivering single core performance and multiple PCIe lanes per socket, ThunderX2 SoCs enable optimized JBOF/FBOF configurations.

Cavium XPliant 3.2T Ethernet switching solutions for NVMe-oF

The Xpliant family of Ethernet switches improves compute and storage application performance in cloud, enterprise, and service provider data centers by offering a variety of features. The XPliant Programmable Architecture enables switching solutions that can be optimized for specific network architectures, including storage networks. Integrated end-to-end network performance management via the Packet Trakker instrumentation tool suite and large fully shared packet buffer guarantee lossless performance required for scaling out NVMe-oF.

The firm’s ThunderX2 ARM64 processors, FastLinQ 100GbE OCP 2.0 NICs with Universal RDMA and XPliant 3.2T programmable Ethernet switches are available from the company and its collaborators worldwide.