Flash storage emerges as viable alternative to spinning disks

Editor's note: This story was changed on April 1 to correct the term multi-level cell, a form of flash storage; and that Nimbus had at one time guaranteed its flash arrays for three years, not three months.

Back in 2011, Lake County, Ill., began testing a virtual desktop infrastructure (VDI), which promised improved disaster recovery, better remote access to county resources and cost savings in operating and managing hardware and applications.

However, when used with the county’s traditional storage-area network with spinning disks, “we were not performing as well as we would like, said the county’s lead IT architect John Clark.

In tests of the business-critical countywide tax system, latency was averaging around 10 milliseconds, with spikes up to 30 to 60 milliseconds. The tax system needed latency of no more than 5 milliseconds.

The delays were traced to logjams occurring because of the increased demands on storage from the VDI environment. This was not completely unexpected, but with the county committed to moving ahead with virtualization of both its desktops and tax system applications, the need for a fix became pressing.  But when opening up additional high-speed channels did not improve performance, about 18 months ago the county began looking into the emerging technology of solid state disk storage.

The timing was good. Flash storage was beginning to move from the exotic to the mainstream, and the county was looking for something that was manageable, easily scalable and “had a good price point,” Clark said. It chose the Gemini All-Flash Array from Nimbus Data Systems and saw an immediate improvement. “We had a six-fold performance increase and our customers are very happy.”

Price versus performance

Although a good price point was important for Lake County, flash storage remains more expensive than traditional disks. It is performance that is driving its adoption, as the demands of virtualized computing run into roadblocks on spinning disk storage systems.

“Disk storage is 40-year-old technology,” said Nimbus CEO Thomas Isakovich. “It’s almost a relic. There’s a demand for something new,” and persistent flash memory is faster than anything mechanical.

Not that flash itself is new. The technology has been around for decades, usually in consumer devices where it has been “good enough” for non-critical uses. But until recently it has not been acceptable for enterprise-level services that require 24/7 reliability with no downtime.

“The thing about flash is that it wears out over time,” Isakovich said. Although it is not mechanical, cells loose the ability to retain an electrical charge over time with repeated writing to memory.

This shortcoming can be mitigated in two ways: Reducing the number of times data is written to a device and improving devices’ ability to retain the charge. Both are being pursued by industry. Single-level cell (SLC) flash can survive 50 times more writes than the traditional multi-level cell (MLC), but it costs 10 times as much. Although this is a good trade-off of performance to price, it still makes it uncompetitive against traditional disks for most uses.

Working around limitations of flash

The second path requires improving data management for flash so that less data has to be written to it. Using this technique, Nimbus now guarantees the reliability of its MLC all-flash arrays for 10 years, compared to just three years previously.

Although this cheaper technology still costs up to nine times more than the least expensive spinning disk array, it now is in line with the most expensive disks, making it an acceptable alternative for performance-oriented applications. Yet for uses such as archival storage and other applications where performance is not critical, traditional disk still has the edge.

Isakovich does not see complete price parity between flash and disk storage occurring within the next five years, but because of flash advantages in power efficiency and capacity, he believes that the cross-over point in affordability could come when the price of flash drops to five times the cost of comparable quality disk storage.

For the present, performance can be a compelling driver. For Lake County, it was not the cost of the technology but the results it produced that was the determining factor in buying the Nimbus all-flash array.

The initial flash array is supporting the county’s virtual desktop infrastructure and the tax system, and the county has purchased a second array to host its Oracle Enterprise Resource Planning software and its Hyperion business intelligence software from Oracle. The county is evaluating all of its database-centric applications to see if it makes sense to virtualize them with all-flash memory.

Enabling a virtual environment does not by itself save money, Clark said; the return on investment comes from better productivity for end users and the ability to maintain systems without increasing the support staff if you can keep storage costs down. The county was looking for a $10 to $100 return on its investment dollar for virtualization, “and that’s what we got,” he said.