Unifying On Premises And Clouds With Flash And NVM-Express

As enterprises continue to embrace hybrid and multicloud models, they are looking for fast access to the data that can be stored both on-premises or in public clouds, and often in more than one cloud. They are also looking for tools that can operate in any environment and can leverage modern storage technologies like flash and NVM-Express protocol.

Several years after first coming onto the scene, NVM-Express continues to get attention from vendors and organizations alike for the promise it holds to ramp up performance and cut into the latency between compute and flash and other non-volatile memory.

Server and storage vendors are driving hard to the hoop to embrace flash storage and NVM-Express as key technologies for data center environments that are expanding beyond the firewalls and into the cloud and out to the edge but that need to be viewed as single entities. After launching in 2009, Pure Storage was heavily focused on all-flash arrays, but soon started seeing NVM-Express on the horizon and began putting pieces into place to bring into the portfolio, including in 2017 launching its FlashArray//X, a revamped flash array that included the vendor’s NVM-Express DirectFlash modules.

Now more than a year later, the company again is rolling out products that leverage flash and NVM-Express to complete a unified on-premises-to-hybrid cloud data protection, access, backup and recovery vision that began coming together in November 2018, when the company announced its Cloud Block Store was now available on the Amazon Web Services (AWS) public cloud. Now the company is adding its ObjectEngine platform for data protection and recovery for both on-premises flash environments and the cloud and the addition of NVM-Express and NVM-Express-over Fabrics (NVM-Express-oF) to the Purity 5.2 engine in its FlashArray//X all-flash storage systems.

The ObjectEngine comes as enterprises are transitioning to what Brian Schwarz, vice president of product management for Pure Storage, tells The Next Platform is an evolving way to look at data protection in the cloud era. Traditional data protection platforms rely on hard drives and tape in what Schwarz describes as D2D2T, where primary disks are backed up by other disk appliances in production sites, and then by tape in secondary sites. Now comes F2F2C, with flash data being backed up by the cloud, which leverages technologies like deduplication and cloud economics to accelerate backup and recovery and reduce costs.

“The slow restore times is a real challenge for people,” Schwarz says. “The cost of these systems is certainly tangible and you’re essentially double buying them for both your first site and your second site, and then you have all the challenges of a tape-based system where essentially the data goes dark. You simply can’t get access to the data once it’s on the tape. The tapes get lost you [or] don’t have the right tape library to reinstantiate old tapes. This new [model is] a massive change that’s going to happen to this architecture in what we call flash-to-flash-to-cloud. We’re at the beginning stages of a complete wholesale change in how people protect their data.”

By using flash as the primary storage and keep the most recent data protect on flash, recovery is accelerated. Then enterprises can really harness “the value of the cloud to both reduce cost because the cloud for archiving is quite compelling and you’re essentially reducing the need to buy a second site and the services that go along with tape archiving and you can also benefit from the use of the cloud processing to basically give yourself the bursting capability to re instantiate that data in the cloud for secondary use cases for virus or malware scanning or to essentially run analytics on the data, etc.,” he says. “You probably don’t want to have all that capacity on your primary premise, but you could essentially use the burst ability and elasticity of the cloud to deliver that. It reduces the cost and gives you that extra benefit.”

ObjectEngine is built largely on the technology Pure Storage inherited in August 2018 when it bought startup StorReduce and its cloud-first deduplication software. The offering comes in two forms, including the ObjectEngine//A for on-premises environments. The four-node, 3U appliance – which will be available in March – comes with deduplication capabilities and delivers up to 25 TB/hour throughput for backup and 15 TB/hour of restore. It offers 15 PB of front-end capacity, supports backup software from Oracle, Veritas, Veeam, and CommVault and connects to the public cloud. It can scale by adding nodes.

Schwarz says the economics of the cloud can drive down the costs for data protection. While enterprise using traditional D2D2T models must populate and manage two different sites, those with the flash-and-cloud platforms keeps backups on the primary production site for a couple of weeks, with the rest of the data sitting in the cloud. That can help make the more modern setup 30 percent to 50 percent less expensive.

For such an appliance, “you needed something that was cloud native, so it’s object storage-centric,” Schwarz says. “You really want it to integrate with existing backup applications. Ripping out backup applications is certainly possible but it’s pretty painful because you’ve a lot of operational procedures, you’ve a lot of clients deployed around your environment and you really want something that with scale out.”

The ObjectEngine//Cloud is cloud-native software for AWS with an S3 interface and a single global namespace. It can scale up to more than 100 TB/hour for backup with a capacity of more then 100 PB in the cloud. It will be available in the second half of the year.

Pure Storage’s DataDirect Fabric is the company’s latest step in its NVM-Express efforts that began in 2015 with the introduction of NVRAM, the first full NVM-Express DirectFlash arrays in 2017 and then the DirectFlash NVM-Express-oF shelf last year. Now the company is adding NVM-Express-oF access into the FlashArray//X offerings. The new DirectFlash provides latency to 250 microseconds, dropping database query time to 2.5 seconds, half the time for iSCSI and 20 percent for Fibre Channel. With NVM-Express-of RoCE (RDMA over Converged Ethernet), enterprises can get flash closer to applications for faster access.

The company brought out NVM-Express-oF RoCE in January and will add Fibre Channel and Ethernet in the future.

The DirectFlash with NVM-Express-oF is giving enterprises an alternative to DAS, according to Chadd Kenney, vice president of products and solutions at Pure Storage.

“There’s a couple other players in this space that are doing already RDMA over Converged Ethernet there now, but none of them have any sort of data services whatsoever,” Kenney says. “But think about matters like a really expensive JBoss and that the challenges with the enterprise is that in order for you to swap out DAS in particular, you’ve got to see a result. Well, those guys typically have the latency results. The problem is that they’re so ridiculously expensive that it’s very hard for people to be able to converge architectures. The good news for a lot of our customers is that they already rely upon FlashRay for a lot of the stuff that they do today and so they’ve seen the resiliency, the efficiencies and the simplicity of the platform. Now it’s just a logical approach to say, ‘Let me get better efficiencies.’”

Hyperscalers like Google and Facebook have been doing this for years, he says, adding that “a lot of them have now gone to complete disaggregation, which is moving compute and storage to completely separate racks altogether. And I think that you know the inherent efficiencies that they get with those are big wins for them, especially at scale. I think we’re excited to finally bring that to the enterprise and bring that hyperscale efficiency umbrella.”

Sign up to our Newsletter

Featuring highlights, analysis, and stories from the week directly from us to your inbox with nothing in between.
Subscribe now

Be the first to comment

Leave a Reply

Your email address will not be published.


This site uses Akismet to reduce spam. Learn how your comment data is processed.