Fusion-io Makes SSDs More Affordable

Solid state drive (SSD) technology is generally expected to grow in importance in enterprise-class storage as the performance tier of choice. In fact, over time, while tier 1 FibreChannel (FC) and serial-attached SCSI (SAS) drives will not be entirely displaced. The SSD market will bifurcate into a performance tier and a capacity tier that will coordinate with each other.

David Hill

July 27, 2009

6 Min Read
Network Computing logo

By David Hill and Jim Handy
Solid state drive (SSD) technology is generally expected to grow in importance in enterprise-class storage as the performance tier of choice. In fact, over time, while tier 1 FibreChannel (FC) and serial-attached SCSI (SAS) drives will not be entirely displaced, tier 0 SSD storage will be used for applications where performance is a paramount objective and tier 2 SATA drives will be used increasingly for applications where price—and not performance—is the primary consideration. That means that the SSD market will bifurcate into a performance tier and a capacity tier that will coordinate with each other.

There is still some question over when this will take place. Would-be users are still trying to figure out how to efficiently optimize SSD.  Many others are trying to understand when it makes sense to adopt this new technology. However, the biggest inhibitor to the adoption of SSDs remains cost. On a per gigabyte basis SSDs are much more expensive (about 20X) than hard disk drives (HDDs).

The economic case for SSDs in the enterprise is instead built using a different measure: Price per IOPS. For example, to achieve very high performance in an HDD-based system, data center managers will often use parallelism, spreading data over many spindles, increasing IOPS but resulting in a lot of unused capacity.  SSDs are inherently much faster than HDDs, so a single unit with less total capacity can effectively replace numerous lightly utilized drives. This is especially true when "short stroking" is applied reducing usable capacity to less than 20% of an HDD for a relatively small IOPS gain.  Usually the single SSD will carry a smaller price tag than the total price of all the HDDs it replaces.

Some argue, nonetheless, that SSD prices are falling so quickly that it is inevitable that they will reach a price-per-gigabyte crossover with enterprise-class HDDs in the near future.  This comes from a misperception about how quickly flash memory prices can fall.  Although flash SSD prices recently fell faster than HDD prices, such falls are not sustainable. Over the long run SSD gigabyte prices will remain about 20X that of an HDD, so any case based on price per gigabyte will continue to pose a challenge. From this perspective price vs. performance will always be the key justification for the use of SSDs in enterprise data centers. Of course, steep price drops will always temporarily improve the business case for SSDs. This is simply Economics 101 and price elasticity.

SSD vendor Fusion-io believes that it has found a way to significantly reduce the price of enterprise-class flash memory devices without reducing  performance or compromising reliability by using what it calls Single-mode Multi-Level Cell (SMLC) technology, which uses multi-level cell (MLC) flash memory technology at the chip level. This compares with standard enterprise-class flash memory which uses single-level cell (SLC). Raw MLC technology is consumer-grade flash memory (which is what you find in removable USB drives) which trades off a much lower cost structure for higher capacity, with as much as 10 times lower endurance than SLC technology.  An MLC chip of a certain capacity can be made on a die that is half as large as a SLC counterpart of similar capacity, so MLC has a significant price advantage. But recently the ratio of SLC prices to MLC prices has widened past the natural 2:1 ratio one would expect.  This is because the volume for MLC has been much larger than for SLC, shrinking the number of SLC suppliers. As a result, the market for SLC has become less competitive and prices have increased to levels as high as six times the prices of their MLC counterparts.

This is a problem for most enterprise SSD makers who contend that only SLC can be used in enterprise SSD applications.  When SLC flash prices spike their SSD prices must keep up.  Fusion-io has been marching to the beat of a different drummer for the past year, now, providing a line of MLC-based drives with specifications that are satisfactory for use in enterprise applications, in addition to their standard line of SLC-based products.

Fusion-io is now tapping into their MLC expertise to go one step further.  SMLC is the company's way of managing MLC chips to provide a drive that matches the endurance and write speed of their SLC offering, but has a slower read speed that matches the performance of their MLC drive.

Since most SSDs provide more read speed than their host systems can accommodate, but are significantly slower when it comes to writes, this trade-off should be very welcome by prospective SSD users, especially since it is being offered at a lower price than are the company's SLC counterparts.

Fusion-io will begin shipping SMLC versions of its ioDrive this quarter.  With SMLC the company is able to create an essentially new "class" of SSD storage.  As in all of its other devices, Fusion-io applies a proprietary blend of techniques in its SMLC ioDrives, such as bad block mapping, Fusion-io's Flashback redundancy and self healing, and error correction coding, to give enterprise-class endurance and reliability, as well.Fusion-io's approach not only presents a challenge to its competitors, but it also increases the potential market through the basic economics of price elasticity. But can this work? History gives two illustrations of how a similar approach succeeded. In the early 1990s, EMC became preeminent in the mainframe disk storage market using SCSI technology with mirroring. Mirroring doubles the price of storage, but still EMC was able to make a compelling economic case and became the leading supplier of mainframe storage in an era when "no one ever got fired for buying IBM." The second case is more recent—the rapid adoption of SATA technology in enterprise storage solutions. SATA drives are not as reliable as FC or SAS drives, but they have proven to be "good enough" for many applications. The ability to use RAID 6 so that two disk failures do not cause a problem if a rebuild occurs before a third disk fails makes the reliability question moot.

Conclusions

Fusion-io has gone against conventional wisdom by introducing an enterprise-class storage device based on less expensive MLC technology, and now with SMLC, is producing a higher-endurance drive with equal reliability and competitive performance at a price point between that of their SLC and MLC-based drives.  In doing so, the company hopes to replicate the success of the two earlier examples we cited: mirroring and SATA. If Fusion-IO's SMLC offerings are successful, the company should be able to accelerate the adoption rate of SSD technology. While competitors are likely to eventually catch up, eroding Fusion-io's competitive advantage, the resultant SSD price war can only be to the benefit of enterprise customers.

Jim Handy is Directors with Objective Analysis.

Read more about:

2009

About the Author(s)

SUBSCRIBE TO OUR NEWSLETTER
Stay informed! Sign up to get expert advice and insight delivered direct to your inbox
More Insights