Non-Volatile Memory express (NVMe) allows hardware and software to fully exploit the capacity of modern SSDs, avoiding bottlenecks -- and shift data even faster.
One of the companies looking to use this is Excelero. Its NVMesh product allows customers to build distributed, high-performance server SAN for mixed application workloads, allowing applications to used pooled NVMe storage devices across a network at local speeds and latencies. Customers include NASA and GE.
ZDNet talked to Excelero's CEO and co-founder, Lior Gal to learn more.
ZDNet: What's the point of difference between you and the competition?
Gal: In 2014 NVMe drives were already assembling. NVMe drives by themselves provide excellent performance - lower latency, better IOPS. You didn't need to be magicians to see that NVMe had a good chance of becoming a standard.
So, what we said was, if you look at the servers - you look at NVMe drives - the future NVMe drive will be one million IOPS, as an example. And in 50 or 40 or 30 microseconds. And the future technologies will talk to very low latency and very high IOPS
If you put that drive or card or whatever, in a server, as long as that sever requires that level of performance or capacity and as long as the application is designed to consume it within the server, then you are getting all the benefits.
But the problems start when you need to share.
And if you try to share, you discover that you are losing those amazing microsecond response times and the mere milliseconds it takes to move the files. And you are losing them by a factor of 10 or a factor of 20 less performance, because you're sharing.
So, what if we build a solution that allows customers to use their usual technologies, with a software stack [not relying on] commodity hardware that would allow them to share n number of servers with n number of drives with the following features:
Number one, it doesn't matter which client accesses which drive. They will get the same performance as if it's local PCI access. In reality today what we are five microseconds over it for remote access which is actually what the network would cost you.
The second piece was, it's ok that I access your drive remotely and I get only five microseconds over it. That's cool, but that's not scalable. So the second design concept is that, based on NVMe technology and based on RDMA technology the latest generation network technology, that can be scaled up these days to hundreds of thousands of clients.
That, for me, as someone who was selling into those sorts of customers, was extremely interesting because it was clear from these guys that they were not going to buy OEM storage from anyone anymore.
It was clear that they were going to buy commodity. So that, for me, was the lesson. How do we come to market in a few years not with 50 features of stuff, but maybe just ten features and we make it so easy to use that people will consume it and take it into production.
That was the market for me: low latency networks and NVMe.
And then there was the sense that everything was merging into the big cloud or people were building big clouds themselves. That was where we had the sense that there was a market here but nobody was going to build for these customers. That's the technical concept that started it all.
Now this is my fourth storage company and what I liked about it was that if I take that solution, and phase one of a solution cannot be fully featured - because it takes ten years to build a fully-featured storage solution - it would need to have a silver bullet to make it suitable for customers.
Now then we looked at the 20 customers I am referring to and the fact is, we had to change the maths. Like just about everybody else, they have the latest and best hardware so if I am going to improve things and come out with something better it is going to be in the software stack.
We are going to sell them scale and with ease of management. What does scale mean when you translate that to business? It means that they can share every NVMe at scale.
If you look at the average consumption of an NVMe drive, you will see between 20 percent and 50 percent utilisation.
Now because they cage those NVMe drives in servers, then if a server requires all of the capacity then, in performance terms, it can be up to 90 percent utilisation. These guys are designing scale out applications but there are many applications where it becomes very hard to define what is the right SKU, the right sizing for 40 servers - to live within the boundaries of the server hardware.
Now, with our experience, we have found that you tend to get from 25 to 50 percent utilisation. If I can mover those customers to between 80 and 90 percent utilisation by using the value of sharing and scale, then we can save them a lot of money.
So instead of buying the same hardware, they can use less and use it in a better way.
The other best part is the ability to manage it better. The management part is on top of that and it manages things for them. I give them performance, I give them better utilisation and now I also add other features. High availability, RAID protection and so on. I have the ability to manage it for data migration, manage it for data migration in a more efficient way, to manage it with the sort of DevOps tools that are available today.
What's the company's goal?
We want to be the leading vendor for cloud deployments, or web scale deployments. The leading vendor for these deployments.
But the point to remember is you look at how the market is predicted to develop from 2016, one out of ten customers will be deploying servers that will be used for storage infrastructure while the other nine out of 10 will still be using OEM storage. So we are not unique. OEM storage can come from many vendors.
But then the prediction is that through the next 10 years, everything is moving into the cloud and companies are building stuff in a very efficient way and a lot of customers are trying the cloud. But a lot of things are going to change and the market is going to flip.
Now the same nine out of 10 servers will servers being used for storage and only one out of ten will be using a server for storage. The market will have completely flipped.
Those same nine out of ten servers will be used for storage and only one out of ten will be OEM storage.
Now look at those customers and see where the market is going. Now it's either going to be those Tier 1 tech giants and anybody who wants to follow them and those followers will be trying to build their stack in an efficient way and for us that's the endpoint because we want those customers to be our customers.
And that's where we have been going - 80 percent of our business focus has been trying to get those guys to partner. Now we have won some huge web scale customers who we are not allowed to talk about.
But that's the DNA, we want to win that market. The strategy for us is to win that market which is why our focus is one performance at scale.
The second leg is that we want to get as many, specialised OEM vendors to sell our solution in the NVMe space.
Exploiting the flexibility of SSDs
Learn more about DBAN ( Darik's Boot and Nuke) the tool that most people who have the odd drive to erase turn to. We think it's both thorough and effective.
Get rid of that old data! One company had "used a service provider to erase and resell old drives but the drives still contained a wealth of highly sensitive information, including user names, home addresses, phone numbers and credit card details.
ZDNet's Robin Harris attended a Hewlett-Packard Enterprise (HPE) event in California where Eduardo Duarte, a Senior Product Manager for media in HPE's 3PAR group, gave an excellent talk on the role he expects storage class memory to take in systems
If you only have a few hard drives to erase using a hammer to pound 6-inch nails through the drives is workable. But if you go through a lot of drives you need something that's a little more professional.
The long-term secular trend in storage is the unbundling of attributes.