All-flash arrays dominate the primary storage headlines, but Reduxio Systems CEO Mark Weiner insists there's still...
an important place for hybrid storage systems that combine solid-state and hard disk drives.
Weiner sees the value of all-flash arrays for "high-end applications where price is not a sensitive issue because it's part of a large project." He anticipates much of the low-end moving to cloud. "And then the midrange is the main business in between," he said.
Reduxio's HX550 Flash Hybrid array targets midrange enterprises. The product features continuous and automatic block tiering that stores data first on flash, inline deduplication and compression, and built-in "BackDating" to enable users to recover data from any point in time. The San Francisco-based storage startup launched its HX550 last year at VMworld, and the product became generally available in November.
In a wide-ranging interview, Weiner discussed the latest trends in flash storage, cloud and copy data management, and how Reduxio is putting a new twist on continuous data protection.
EMC called 2016 the year of all-flash for primary storage. What's your take on those that claim everyone will use flash for primary storage, or production data?
Mark Weiner: I just think they're wrong. As an example, let's start at the lower end. I'm pretty sure that all the cloud providers and object stores will not go all-flash. It would be completely prohibitively expensive and not necessary.
An X-ray in a hospital is pretty important. You wouldn't want anybody to lose your X-ray. But if that's on disk drives on object storage that's been centralized for the hospital complex, you're pretty happy with that as long as you can get it in a couple of seconds.
If your only application is watching videos, tape is a great solution still. There's just no reason for [flash]. It's nice. Wow, it's all-flash. I like flying a private jet just as much as the next guy. But it's not for every application.
There's room for different technologies. There's going to be different classes of service. At the end of the day, you as a client just want to have your data reside on the correct class.
I would say "all-anything" is dead. Today, all these flash vendors talk about all-flash. Well, what's going to happen in a year, a year and a half, when, for example, new technologies like memory channel become mainstream or [3D] XPoint, which is being pushed by Intel and is much faster with a much lower latency than flash? Does that mean they're not going to use this technology within their arrays? They're going to remain with old all-flash?
You mentioned EMC. I saw the Unity announcement, and it's nice. Part of the Unity announcement is an all-flash midrange [array], which is more power to them. But the Unity announcement also emphasizes hybrids. The fact that there's a Unity hybrid only validates what we're doing.
Mark WeinerCEO, Reduxio Systems
Do you think the midmarket is more likely to use hybrid arrays?
Weiner: The midmarket will always have a mix of technologies, and the people who are better able to manage that mix of technologies and deliver it to the client in the simplest way are going to be the winners.
If in two years flash prices come down enough that it makes sense for the midmarket, I think flash will penetrate the midmarket entirely. On the other hand, does that mean we can't have a hybrid that, for example, might have different types of flash in it? Or, might be a mixture of Intel's XPoint technology or NVMe drives and 3D NAND?
There are always going to be faster, expensive ways to store things and cheaper and slower ways to store things. It's physics. And I think the people that can utilize these together are going to have faster, cheaper machines for the midrange. Everyone else who is limited to an all-something technology is always going to suffer. You see that today. The reason all the mainstream vendors had to go buy flash companies is because they were based on all-disk drive technology.
Do you agree with vendors that claim all-flash arrays can be equivalent in cost to disk, after factoring in deduplication and compression?
Weiner: It only works if nobody else has the dedupe technology. We have the same dedupe technology. So, they're still five times our price. If you take the all-flash vendors and you factor in their compression and dedupe, of course it makes their boxes cheaper. But they're still very expensive, even at that level.
Reduxio's hybrid array is 25% flash and 75% disk. Is it possible you'll tweak the amount of flash, or is 25% the right amount?
Weiner: I don't know. I'd always like to go less. I don't think we need to go more. For all I know, it could be far less, but we don't want to do that yet.
Let's say our box stretches to about 100 TB. Say 25 TB is SSD. Right now, that definitely works, and we're seeing amazing performance. Could it have been 20 TB -- 20%? Could it have been 15%? Maybe. But we don't know that yet.
For now, it probably doesn't matter because it doesn't change the cost so much at that small level. But where it could matter a lot is when we get into the next generation of technologies, like memory channel, like XPoint. Then it will be very interesting because if you can use a small amount of one tier, that could boost your performance enormously.
Does Reduxio's BackDating differ in any way from the continuous data protection (CDP) technology that's out there?
Weiner: We abstracted all the data and put all blocks in a database. The database does the compression. The database does the dedupe. The database keeps track of every block and what time it hit the system. So when you look at a file, all we're doing is presenting the data by doing a search of all the blocks required in that data center now. When you want to look at your data that was five seconds old, five minutes old or two days, three hours and 30 seconds old, we simply do a search on the block with that time stamp.
Basically looking at the data now or looking at data from any past point in time, for us, is the same operation. It's CDP where you don't require external equipment. You don't buy anything else, and it just works.
CDP came out 10, 15 years ago. It was a very cool concept. But the challenge with CDP, it was always a secondary storage. You already had your primary EMC or what not, and then you had your secondary storage, where something was splitting the data stream and creating a continuous copy on some other device. You had to buy all of this secondary equipment.
It's almost like this whole business of copy data management. You begin by explaining, "We're just going to eliminate all the copies of your data. We're going to reduce the complexity. We're going to have an infinite history of your system." And you're thinking this is great. This is going to save me a lot of money. And then the first thing the sales guy suggests is buying a whole bunch of new equipment and making more copies. And you're thinking, "Wait a second. That doesn't make sense. You told me I was going to eliminate my copies. Now you're telling me I've got to buy more and make more."
Copy data management's a cool concept, but it should be on the primary device. Otherwise, you end up buying a whole lot of new equipment. And you add a whole lot of cost and complexity to the organization when the goal was actually the opposite.
What's your view on cloud storage?
Weiner: We love cloud. We hook onto the cloud. We are going to be part of a cloud infrastructure. We view cloud as another tier. We think that the world will work in many tiers, and you've got to be able to take advantage of this.
Let's talk about what cloud is. Cloud is write once, read many, higher latency, lower cost, large storage. Let's not say cloud is just Amazon. It could be a private cloud you set up with one of the software vendors like [IBM's] Cleversafe or Scality.
In our world, we have your fastest data sitting on the fastest media, and when you need to grow, we'll grow onto other media. In other people's worlds, let's say an all-flash vendor, when they have one more megabyte than the size of their array, they today have one answer: Buy another array and cluster it. I don't think that's a particularly good answer. That's a particularly expensive answer.
Reduxio appears in our list of storage startups to watch in 2016
All-flash arrays make inroads against hybrids
The skinny on what you need to know about hybrid storage arrays