IBM Makes Economic Case for All-Flash Arrays

In its recently announced flash initiative, IBM came to what at first may seem to be a startling conclusion: that all enterprise-class Tier 1 (i.e., high performance) storage should be flash. Let's not start a "disk is dead" movement, however, as disk will continue to play a major role in storage, such as for Tier 2 capacity storage.

Observers have predicted the end of hard disk Tier 1 storage for some time, but few expected it to happen within the next few years. While SSDs in the form of non-volatile flash memory have dramatic and well-touted performance advantages, their cost per GB of storage has been far higher than the best-performing hard drives. Conventional wisdom suggested that a good economic case could be made for using flash-based products only for that portion of storage that truly required high performance (and that was considered to be only around 5% to 20% of Tier 1 storage).

In contrast, IBM contends the economics favor flash over hard disks. How can that be? Most previous analyses compared flash memory with hard disks. When storage administrators use hard disks for performance applications, they typically only load a fraction of the data that drives are designed to hold to get the necessary IOPS. Because flash has no mechanical constraints, it can deliver similar or better performance in spite of being fully loaded.

IBM has taken a different tack. Computing is not just about servers alone, but also about the networking and storage resources from a physical perspective and all the software, including applications, that have to be put together to create a "system." From IBM's system perspective, flash uses less physical space and fewer network connections, which will impact and improve overall performance. Let's examine some of the other considerations from the example that IBM presented.

• 17% fewer servers -- Fewer servers mean fewer cores and fewer network connections, and also lower power requirements.

• 74% lower environmental costs -- Flash takes up much less floor and rack space, as well as being much more environmentally (and cost) friendly in terms of power and cooling.

• 35% lower operational costs -- IBM calculated the savings in server/storage administration (people time).

• Higher storage utilization -- While flash still costs more, 50% better storage utilization, lower maintenance and simplified management helps to ameliorate the difference.

[ Join us at Interop Las Vegas for access to 125+ IT sessions and 300+ exhibiting companies. Register today! ]

Now, all these are important reasons, but the cost for flash was still about $2.1 million versus about $1.7 million for disk. While $400,000 or so dollars is nothing to sneeze at, it wasn't the compelling factor that tipped the scales dramatically:

• 38% lower software license costs -- Software is the cost tail that wags the hardware dog; lower cores and other changes drive lower costs for database and infrastructure software. This is dramatic. In the example, IBM showed costs fell from about $5M to $3.2M; these savings of $1.8M dwarf the $0.4M.

Overall the cost with all disk was $7.1M, while the cost with all flash was $4.9M. This $2.2M savings meant a savings of 30%. In IBM's view, the winner and new champion is all-flash.

Next page: But What About ...Still there are some possible concerns that have to be addressed in IBM's analysis.

• Realistic examples -- IBM stated that its analysis was representative of a common situation and would apply on a general basis within an enterprise deployment. Its calculations may not produce the same results in small businesses and mid-sized organizations.

• Scalability -- The ability to scale means that you have to be able to climb the 1 PB or more storage mountain. IBM confirmed that it has at least one customer with greater than 1 PB of flash storage.

• Reliability and durability -- Flash memory is known for wearing out after so many writes, which is a legitimate concern for organizations that plan to invest significant sums of money. There are workarounds, of course. For instance, software designed to help lessen or avoid the problem is deployed nearly universally; and manufacturers warranty their products for five years. Of course, redeeming a warranty may not be sufficient compensation if a flash failure has a significant impact on your business or requires a great effort from staff to recover.

• True commitment -- A customer who buys into the flash model wants to know it has a vendor partner that's also committed. So how committed is IBM to all-flash? It has promised $1B for flash research and development; it has set up 12 Flash Centers of Competence around the world, and it has announced the IBM FlashSystem that builds upon its acquisition of Texas Memory Systems (TMS); IBM is committed.

No Instant Transition to All Flash

IBM competitors in the all-flash array space (from large players, such as EMC and NetApp to smaller players, such as Kaminario, Nimbus Data, SolidFire, and WhipTail) are probably thrilled with this announcement. IBM's commitment may help loosen up the market, and these competitors can point to IBM's analysis that all-flash arrays are economically justifiable for replacing all Tier 1 storage when they talk to customers.

Still, the transition will not occur overnight. First, unless there is a compelling reason otherwise, businesses will continue refreshing their Tier 1 storage in roughly three to five years. Second, enterprises will have to kick the all-flash storage array tires, and that can take some time. And don't forget uneven adoption cycles in which many are slow to adopt new technologies (or may delay adoption indefinitely) although there have already been a lot of flash deployments.

Finally, there is a wild card factor. A wind that is good for some (all-flash array companies) blows ill for others (software companies). Remember that the majority of the savings in IBM's analysis comes out of the pockets of software companies whose traditional licensing models are built around traditional architectures.

Those vendors will probably adjust their license models to try to extract more money from all-flash array environments, as well as create their own hardware environments with the claim that everything runs better as it is all from us. This is not likely to be a totally successful effort (as they don't want to cannibalize existing customer revenues, and changing licensing models is not easy), but they are likely to react if they see their sales affected.

Mesabi Musings

Just as tape storage is not dead, so the appearance of IBM's new Flash initiative and investment does not mean that disk is dead. But similar to tape, disk's role is changing. Use cases, such as active archiving for large capacity requirements, are likely to continue. This means large-capacity disks are likely to continue serving useful purposes. But Tier 1 hard disk drives' days are numbered (although somebody will probably find some specialized use cases for them).

In the meantime, we can expect to be inundated by all-flash array vendors as they seek to gain market share. IT customers should benefit from the competition, but they should also plan a qualification and adoption strategy that allows them to enjoy and profit from these innovations at a pace with which they are comfortable.

IBM is a client of David Hill and the Mesabi Group.