How QF2 on all-flash delivers the best storage performance

Posted .

 
3card-ds-pseries

QF2 powered by all-flash data sheet

This two-page PDF data sheet gives an overview of QF2 running on all-flash instances. Download to find out the features and benefits of QF2 on all-flash.

Download now
3card-promise

White paper: Promise of Universal Scale

QF2 is the first universal-scale file system, spanning on-premises and the public cloud. Download the white paper to find out more about this new category of storage.

Download
3card-biren

Qumulo blog: QF2 for all-flash

Visit our blog for more on QF2 for all-flash, where Biren discusses some of the ideas from the webinar.

Read More

Transcription of webinar

Hello, everybody. I am super excited to announce that we will now have Qumulo File Fabric on an all-flash hardware system. And we've been watching the market for quite a few years, and we've seen a whole bunch of market trends that make it the right time to launch our File Fabric platform on all-flash. So let me talk to you a little bit about what's changed in the market over time.

There were several industry verticals where performance requirements have skyrocketed in recent years. Take, for instance, like life sciences and medical research. Genomic sequencing has been rapidly becoming the norm on how researchers actually try to diagnose and treat diseases and come up with new medication to cure new diseases. And if you think about the genomic sequencing workload, it's not just about how much performance that workload needs, but it's also the access patterns.

Researchers try to match sequences to reference data sets, and the access patterns are often very random in terms of access. And it's a workload that is not easily addressed with spinning discs that do not have the performance profile of an all-flash system. You can see such trends in several other verticals that are on the screen. I'm not going to touch upon every one of them, but if you think about media and entertainment, a couple years ago we started thinking about 4K video, and now we talk in 6K, 8K video streams, and many of these video streams are monolithic. Some of them are actually frame-based.

Again, there is a very exacting need in terms of how fast these frames need to be served, especially if you're talking about real-time video screening and real-time 4K or 8K video edit. So again, we've seen a whole bunch of interest amongst customers and prospects in the media and entertainment space because of that growing need and that in that use case. Artificial intelligence and machine learning is something that most of you must be familiar with. Autonomous driving vehicles do a lot of data access to do image recognition, when they actually have to decide how to steer a vehicle on a road. So there's a growing interest amongst that set of consumers, as well, for high-performance storage systems.

So moving onto the next slide, why is QF2 any different than what other offers might be out there? If you think about other file storage systems that support all-flash, you often find that they have three big limitations. First thing that most other players in this segment do is that they tie you to their own proprietary hardware. Instead of actually focusing on what matters to consumers, which is high-performance, having a software system that's scalable, reliable, and protects your data, they actually expend energy and effort in building highly custom hardware components that require their own expensive supply chain. And what's even worse is that these custom hardware products become obsolete in 24 months. It's not standard hardware. It's something that's very, very specific to the vendor that you're buying the store the solution from.

The second problem with that, which is that these vendors do not have the scale and breadth of people who are experts in providing hardware solutions. So instead of focusing on what they claim to be very good at, they're actually expending energy not just printing software but also building the hardware components. What happens, then, is that you often end up with very, very custom components. It could be a custom device driver or it could be other low-level system components. But the point here is that you're getting locked in into stuff that is not custom and because these vendors don't have the wherewithal to actually have rapid innovation because they only have so much a budget to actually expand towards software improvements as well as hardware, you don't get to ride the standard hardware improvement trends that happen over time with vendors who actually specialize in creating hardware, and it's their bread and butter.

The last thing that happens when you get locked into a custom hardware storage solution is that your data's stuck in a walled garden. You are running a storage system with your data in it, and it can only run on that storage system. So data mobility becomes an issue. Take, for example, the growing need amongst many prospective customers where they wanna actually move data from their on-prem storage system to the cloud to do some burst computing because they want an elastic capacity because they're running low on capacity during peak season. That becomes a challenge when your entire file system is dependent on custom hardware that's sitting on your dataset.

So our intent in launching QF2...not just for all-flash, but on all the platforms that we support...is to divorced platform from what do you really care about, which is your data, and you have it where you want it quickly and be able to access it quickly. So then the question is, "Why now?" We have been in the business for a few years. There have been other players in the market that have actually launched file systems using custom hardware that I just alluded to in the past year or so. So why do we think that this is the right time for us to launch and all-flash system? To answer that question, I think there's a very straightforward answer. Everything we've done, everything we do day in and day out at Qumulo, is based on what our customers say.

In fact, one of our leadership principles is that customers are our magnetic field. In everything we do, we ask ourselves the question if whether or not that would actually add customer value. And we've conducted hundreds of customer interviews for many features...and not limited to just all-flash, but even other features that we built. But specifically for all-flash, we've at least conducted 50 plus customer prospect interviews to see whether this is the right time we should be launching an all-flash product, and if so, what the product should look like.

And when we started Qumulo a few years ago, what's not very well-known is that our first product was actually an all-flash system. So we were actually first to bring an all-flash system that supported a scalable file system to the marketplace way before the competition. And our Qumulo File Fabric was built keeping an all-flash system in mind. But then we shifted gears, and we went the route of actually morphing our File Fabric to support a hybrid system, where the flash tier was mainly for hard data and to serve data needs that had extremely high-performance needs. And hen the backing hard disk tier was intended to store data that did not require rapid access. Why did we do that? Well, the reason we did that is because our customers at that point were not really interested in flash systems because they were extremely expensive. Flash, at that time, was 7 to 10 times more expensive on a per-terabyte basis compared to spinning disks.

So a lot of customers told this, "We really love your system. It's very fast, but we really like the economics of spinning disks to bring the price point down to a more reasonable amount." So we invested energy and effort to move away from an all-flash system to a hybrid system and built our hybrid product portfolio that we've been selling for the last couple of years. And then, instead of actually focusing on building on all-flash system that was not economical for most of our customers that time, unlike our competition, what we chose to do instead was invest in our software performance.

So I'm going to talk about that for a minute here. So over the past year, we've done a lot of work to improve how our software performs on our hybrid platforms. For example, a year ago, our single-stream reads were about a third about what they are right now. We've increased single-stream reads by a factor of 3x in the last 12 months. Our multi-stream reads has gone up by 166%. Multi-stream writes, we've actually increased by a factor of 140%. And SpecSFS, which is an industry standard for measuring IOPS, we have actually increased it by 70%. So we knew that at some point...for two reasons. We knew that customers would appreciate higher performance from their hybrid clusters that we had sold them. But more importantly, we knew that all of these improvements would pay dividends when we actually launch an all-flash system that we always knew we had to launch at some point, when the timing was right.

Moving onto the next slide, there's another reason I think the timing is right, which is the fact that media prices have trended such that all-flash costs are very, very close...in fact, converging...to spinning disk prices. What you see on the screen right now is the dollars per gigabyte of NVMe pricing over time...that's the red line...compared to SATA drives on a dollars-per-gigabyte basis. So if you look at, say, January of 2016, about two years ago, the pricing of NVMe drives was almost double that of SATA drives. And I'm comparing NVMe to SATA SSDs, really, not spinning disks.

So instead of just rushing into the market and building something that is all-flash but using old technology that is SATA-based, we decided to watch what the next generation of flash is all about, which is NVMe. And I'm really excited to say that what we've seen in the marketplace today is that NVMe SSDs are almost at price parity with SATA SSDs. What that means is customers don't have to have any compromise in terms of what media they have their boxes. We have waited for the market trends to change such that we can build our all-flash platform using cutting-edge, next-gen technology, and not using media from a decade or two decades ago. So that's the other reason I firmly believe that now is the best time you could actually launch QF2 on an all-flash platform.

So let's look at how that looks. QF2 is going to deliver the best price performance of any all-flash system out there for file systems, period. QF2, on a per-node basis, can deliver is screaming 16 gigabytes a second of throughput, and it can scale up to 400 nodes for a total throughput of 1.6 terabytes a second. There's nobody else out there today that does that. It's also the first and only all-flash system that runs on modern, standard hardware. We do not build a single piece of the component that goes into that all-flash box that they're gonna support. Everything is standard hardware. Everything is latest, cutting-edge technology, but nothing in there is specific to our company. It's something that customers can actually buy out in the market.

Last but not least, your data is not going to be tied through an on-prem system and reside in a silo. QF2 is a File Fabric that goes across hybrid systems, all-flash systems, and cloud, so your data gets unprecedented mobility. You can move the data where you want. You can stream to the cloud, replicate data to the cloud, compute it if you're running low on on-prem capacity, and bring the results back with on-prem Qumulo cluster seamlessly. That's another advantage that we believe differentiates us from other offerings in the marketplace.

So how does the SKU look? So the first all-flash platform that we'll be launching is gonna be a 2U chassis. It's gonna be a 24-bay, 2U chassis. It's gonna have 192 gigabytes of RAM, and we're gonna support it with two NIC configurations. We're gonna have if a dual-40-gig NIC version, and then we're also gonna have the same box with a dual-100-gig NIC variant for customers who really want more performance than the dual-40-gig NIC can deliver. We're gonna have the same chassis, but we're gonna support two SKU variants. We're gonna have a 23-terabyte 2U which is gonna be a sparsely populated chassis, where only 12 of the 24 bays have SSDs in them. It's gonna be 12 1.92 SSDs. And we're also going to have a denser 2U SKU that's gonna have 24 3.84-terabyte SSDs in them, for a total raw capacity of 92 terabytes. So when this goes on sale, you're gonna have a 23-terabyte 2U option and a 92-terabyte 2U option. Both SKUs will have a minimum cluster configuration of 4 nodes, and you can scale it up as needed. But the minimum cluster size will be 4 nodes.

That's all I have today. Again, we're super excited to bring QF2 and all-flash for our customers. Please stay tuned. You're gonna hear more about our product and QF2 in the coming weeks. I thank you for your time, and we look forward to hearing your feedback on this presentation. Thank you.

Let's start a conversation

We are always looking for new challenges in enterprise storage. Drop us a line and we will be in touch.

Contact Information

REACH US

EMAIL

General: info@qumulo.com
PR & Media: pr@qumulo.com

WORK WITH US

SUPPORT

Search

Enter a search term below