Boosting Enterprise Application Performance in a Flash

Not all Flash memory systems are created equal. That’s the message from the folks at Marvell, which today at the Flash Memory Summit announced the general availability of a DragonFly adapter for servers starting next month that combines caching software and solid-state drives to dramatically boost application performance.

The Marvell DragonFly platform is based on a system-on-a-chip (SOC) architecture that utilizes ARM processors to offload the management of Flash memory from the server. According to Shawn Kung, director of product marketing for the Storage Business Group at Marvell, this approach is necessary because it makes sure that the Flash memory platform is application-neutral, while taking advantage of Marvell software and ARM processors to dramatically improve both read and write I/O performance.

Marvell DragonFly is delivered as a PCIe Gen2 x8 adapter with up to 8GB SODIMM ECC DRAM and up to 1.5 TB of external SSD storage. Kung says this approach will allow IT organizations to significantly reduce dependencies on expensive disk storage, while at the same time reducing the amount of power that those systems would consume in the data center. Flash memory systems won’t completely eliminate the need for disks as primary storage overnight. But it does make it more feasible for organizations to store data on low-cost SAS drives for persistent long-term storage, while relying on DRAM memory on processors, caching software and SSDs to maximize I/O performance.

Kung says that Marvell’s approach differs from other offerings in that the company is not just throwing consumer-grade SSDs at enterprise-class I/O problems. The DragonFly caching software, says Kung, allows IT organizations to take advantage of consumer-grade SSDs that now cost 80 to 90 cents per gigabyte in a way that provides 3.2GB/s of throughput; less than 22us latency; 220K IOPS read; 220K IOPS write; integrated ultracapacitors to protect data in the event of power loss; and support for variety of operating systems, including RHEL, KVM, Xen, VMware and Windows.

Obviously, interest in all-things Flash memory is on the rise all across the enterprise, with the core issue being how to sustain application performance in a Big Data era characterized by virtual machines that generally have significant latency issues that need to be addressed when running mission-critical applications. To what degree those applications will need to run in memory or simply rely on in-memory databases remains to be seen. But the one thing that is clear is that there will soon be a lot less reliance on disks as a mechanism for processing I/O requests from those applications.