Casting a wary eye at the heated competition among the upstart solid-state array vendors, Hitachi Data Systems...
(HDS) Corp. claimed to sell 26.5 PB of flash storage during the last quarter through its major product lines.
Roberto Basilio, vice president of product management at HDS, said flash use is growing among the company's enterprise customer base, and the greatest volume sells in hybrid arrays that offer tiered storage with hard-disk drives (HDDs) and solid-state drives (SSDs).
Basilio estimated that 50% of Hitachi's Virtual Storage Platform (VSP)/VSP G1000 arrays and 35% of the smaller, modular Hitachi Unified Storage (HUS) VMs sell with some flash installed. He said about 10% of the HUS VM units and 5% to 6% of the VSP/G1000 arrays ship in all-flash configurations. The VSP G1000, announced in April, is the follow-on to the VSP array.
Market research reports sometimes exclude Hitachi's all-flash configurations because they trace their origins to the company's HDD-based, general-purpose storage, in contrast to the wave of new arrays designed exclusively for flash. HDS has countered that it devotes considerable effort to optimizing its products for flash, including building HDS flash module drives and making flash-specific modifications to the operating system.
"HDS is not this ghost in the market," Basilio said. "We actually are relevant, and we are actually doing quite a lot of business."
In an interview with TechTarget, Basilio discussed customer trends, HDS flash competitors, the company's plans for inline deduplication and compression, and future possibilities for flash tiering.
Have you seen an escalation in demand for all-flash arrays?
Roberto Basilio: Yes, we do see interest in the all-flash array but it's mostly in the small form factor. Enterprise is becoming more and more confident in the need to use flash in certain types of applications -- [business intelligence] BI and VMware and so on. And clearly they are taking advantage of the falling cost of flash.
But, very, very few need all-flash. There are specific applications that would benefit a lot [from] the smaller all-flash arrays that you see in the market right now. For larger transactional systems, the hybrid form is still the preferred one.
In general, enterprise customers are quite happy with what they get off of the hybrid system. They have the performance. They have the low latency and response time. But, at the same time, they have the benefit of tiering.
Cost remains a factor obviously. We are talking about still a substantial multiplier where [flash is] six to eight to 10 times the cost of disk. So, not every budget can afford that. That is the reason why no one is really going all-flash, and that's reality. However, the prices are coming down. There are different types of flash available as well. Some are more affordable than others, but they are also clearly with different degrees of reliability. So, they are not suitable for all kinds of applications.
Which vendors do you view as your top competition in the hybrid and all-flash spaces?
Basilio: In the hybrid space, without doubt it's EMC. The sheer power of their sales force makes the competition stronger. In the pure flash space, the more aggressive one with more success is actually IBM, but that doesn't mean that they are actually the leader. The top leaders right now are the likes of Pure Storage and SolidFire and so on. Those are the one that are getting more market attention. But, from a volume and really pure business aggressiveness [perspective], IBM with the [Texas Memory System] TMS line [that the company acquired] is actually the one that is more competitive against us.
Do you sense that the type of enterprises that used your traditional storage systems are more comfortable dealing with established vendors?
Basilio: There is also a completeness of the product offering, because it's not just enough to be fast. If you are an enterprise, that data has quite a substantial meaning in your business. You need to be able to protect, to provision, to create a disaster recovery situation and so on. If you take for example our G1000, I can put flash on both systems. I can have that system replicate with one single I/O to two systems … one fail to the other, one continue, so there is no application corruption at all. I can migrate from one system to the other without disruption. These are characteristics that the enterprise workloads require, and the smaller players do not offer that today. I say today because clearly everything can evolve. But, it would take time. It took 20, 30 years for us to get to this level.
When you ship arrays in an all-flash configuration, are customers using them for multiple applications or to accelerate a single application?
Basilio: It's mostly multiple applications. We are the king in the [online transaction processing] OLTP space, so multiple streams of high-demanding transactions. Single application is probably best served from a cost perspective by smaller units like the ones offered by other companies like Pure Storage, SolidFire and so on, even [EMC's] XtremIO.
All-flash array vendors have been adding more and more enterprise features and capabilities to their products.
Basilio: The only things they are bringing that are not yet in our systems are dedupe and compression. We have some compression, but that is at the flash module itself. It's not at the system level yet. There is not yet deduplication. However, we have a file system that we can put in front of our storage, and in that sense we can take advantage of inline deduplication. But, the block alone, no, it doesn't have deduplication.
It that something you intend to support?
Basilio: Yes. We are definitely working on bringing dedupe at the block level. And, with the next version of our flash device, you will have even better compression. Dedupe and compression are two things that are used quite a lot in the enterprise because they lower the effective cost of the space.
By the way, there is a big gap between what is claimed and what is actually achieved. I was recently with a telecommunications company. There was clearly a big laugh when we were comparing the advertised versus the effective [data reduction ratio]. I know there are three, four times the gap between the two numbers.
Do you expect inline deduplication and compression to be available next year? How important is it to HDS flash strategy?
Basilio: Yes, next year. It's critical from a cost performance [perspective]. For those that are very cost-sensitive, it's obviously something that we need to add to our offering. And the market is getting bigger, and there is even more opportunity to go outside what we do better today. So, we're going to go after those VDI environments, those virtualized systems and so on. We do have the knowledge and the know-how, and we can push ourselves in those spaces as well.
What's the largest all-flash configuration you have sold to date?
Basilio: It's in one of the most famous Internet companies, and we are talking about probably a few hundred terabytes per system. It's on a VSP configuration, and the volume of workload that runs on those systems is unbelievable. If I recall, it's 50 billion transactions per day or something like that. And, for them, it's critical to be always on because every second is worth $7,000 for them.
Are they running transaction systems?
Basilio: Yes, and by the way, they are not unique. We have a very large site in the north of Europe, a financial institution. They use in excess of 100 TB.
So, there are some large all-flash installations. But, when you think about the trajectory going forward, do you foresee flash cutting into hybrid sales? Or, do you envision hybrid having the greater growth rate and perhaps disk usage shrinking?
Basilio: I think, in general, we should say that flash as a medium is taking more and more out of the performance disks. That is what we are seeing. Will disk disappear? Not in our lifetimes. It's still a much more economical medium compared to flash.
But, there are also other new technologies. We cannot foresee what would be the actual medium in five to 10 years. Flash came out just six years ago. We are talking about 2007, 2008 when we first started to introduce solid-state devices into our disk subsystems. And flash has made quite a lot of progress, but it's not so fast.
In the meantime, we have other types of technology. Memristor has changed and so on. That could be the medium within the next five to 10 years as well. What I always like to say is that memory-based [technologies] will become more and more relevant for the primary workloads. For the [storage] of data for the long term, disk drives will always be more cost-effective in the long run.
We are talking about already 16, 20 TB per disk in the next 12 to 24 months. So, you see the gap that exists in terms of capacity and cost. There will be these two worlds. There will be the high-capacity disks where you keep the information that you need and not necessarily the one that you access quite intensively. And then there will be flash or memory offerings that will range from the write-intensive to the read-intensive to the very, very cheap ones. But clearly, each one will have a declining level of reliability in line with the technology.
HDS makes its own flash module drives, unlike many array vendors. Do you think that still provides an advantage?
Basilio: A customer tried and compared the two. They found a substantial difference between our technology and the one off the shelf, and not only in terms of performance. … They really last longer. They perform longer. Most of the devices are very fast for the first few hours and then start to go around and do garbage collection. They're trying to keep up with this pace and so on. We are much more intelligent and faster because of the way we built them.
Do you expect to continue making the flash modules?
Basilio: Absolutely. You will see the second generation coming out in a few months, even faster and with even more and better space management. There will be an improved compression algorithm in there to increase even more the space that we can effectively provide.
Some vendors still use single-level cell and enterprise multi-level cell (MLC). Hitachi uses MLC flash. Triple-level cell and 3D NAND are emerging. Do you foresee tiering between different types of flash?
Basilio: I see different tiers of flash being used -- one that is more for write-intensive type of operation, [another] that will be more for read-intensive, [yet another] that could be the primary storage … one that could be the secondary tiers, which you normally tend to read more than write, and so on. I don't want to say what we are looking at right now, but definitely we are looking at a different level of flash to be used within an automated system.