Marvell sees new DragonFly platform as boon to channel partners

 Marvell sees new DragonFly platform as boon to channel partners

Silicon solution vendor Marvell has announced the general availability of the Marvell DragonFly platform, featuring DragonFly NVRAM and DragonFly NVCACHE. Aimed at the enterprise storage market, DragonFly is the storage industry’s first intelligent NVRAM SSD accelerator — essentially a low-latency write-back cache running embedded software in a PCIe adapter form factor.This second iteration of DragonFly, which was announced back in April, underwent some low level tweaks to the architecture in beta, and it now only going to general availability. But the company thinks it is worth the wait, and offers a clear advantage over software cache accelerators.

“Our value proposition is very unique, and very different from the software side,” said Shawn Kung, Director of Product Marketing for Enterprise Storage at Marvell. “Software-based caching has issues and is not that efficient. It is very tightly coupled with the OS, so if you have different OS or kernels, which is common in the enterprise, a lot of code needs to be migrated and that creates risk.” Kung said DragonFly overcomes that problem.

“All the intelligence runs on our ARM-based processor, which decouples you from the OS for a riskless method of supporting multiple OS,” he said. “All that code is pushed down, and you also don’t have to worry about the CPU and memory resources up on the host.”

Kung also said DragonFly makes high-performance caching much less complex, delivering the highest IOPS and lowest latency performance in the market which avoiding a potpourri of expensive DRAM, capacitors, SSDs, generic software and custom professional services.

“This is ideally suited for customers who have a pain point on scaling infrastructure without storage IO or latency being a bottleneck, who want predictable, consistent read-write IOPS and low latency,” on random write data,” Kung said. Typical customers will be using database applications, virtualization, or loud infrastructure.

Marvell DragonFly NVCACHE leverages the industry’s first hardware-assisted write-back caching on tiered non-volatile memory (NVRAM + off-the-shelf SSDs) to yield 10-100x higher virtual machine IOPS, while lowering NAS/SAN and local DAS IOPS demands by 50 percent or more.

The first in a new category of NVRAM-powered write-back caching solutions, the DragonFly platform plugs directly into commercially available servers with a PCIe slot. Combining Marvell’s System-on-a-Chip technology with its newly designed circuit boards delivers a leading-edge sub-system solution that can consistently achieve 10-100x improvements in latency and server I/O performance while reducing power, space and storage capital costs in the datacenter. DragonFly is sold as a PCIe Gen2 x8 adapter with up to 8GB SODIMM ECC DRAM and up to 1.5 TB of external SSD storage, It also offers read, write-back and write-thru caching, synchronous low-latency peer-to-peer write-back mirroring, virtual machine migration and management tools, 3.2GB/s throughput, less than 22us latency, sustained 4K random performance of 220K IOPS read and 220K IOPS write, integrated onboard ultracapacitors to protect data in the event of power loss, and a wide variety of host operating system support, including RHEL, KVM, Xen, VMWare and Windows.

All of this will offer significant new opportunities for Marvell’s channel partners, Kung said.

“The hard drive and SSD business at Marvell are direct, but storage products are 90% channel and our PCIe controllers are channel,” Kung said. “We are working with system integrator and VAR partners on this, and have signed on new partners both in the US and in Asia Pacific that have sold and distributed board level products. We really want to embrace the channel going forward.”

Kung said DragonFly offers partners benefits that other SSD accelerators don’t have.

“The whole point of the Dragonfly product is you take an adapter that’s turnkey and drop in and it works, and you can attach consumer grade quantity SSD off the shelf, and you can create very high IOPS and latency performance,” Kung said. “There’s no need to pay $5,000 for a high end PCIe card. And the channel likes that.”


Leave a Reply

Your email address will not be published. Required fields are marked *


82 + = 84

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>