Does software-defined storage stack up?

For most IT buyers, storage is a bit like air: it’s not important unless you’re running out.

But as data explodes and storage is sucked up, capacity is increasingly choked. Managing storage is suddenly very important and it’s a golden opportunity for storage resellers to provide a breath of fresh air.

Virtualised storage and solid-state media are now revitalising a field of computing that has consistently relied on the spinning disk since it abandoned tapes 30 years ago. Software-defined storage (SDS) is becoming especially important in a new world reliant on hybrid and private clouds, with scenarios demanding a mix of spinning disk and flash media.

But nothing in this industry is straightforward. Similar to software-defined networking, the precise definition of SDS varies according to each vendor’s interpretation.

To add to the confusion, storage is a hardware component that traditionally has had significant software control placed over it, with controllers such as file systems, data management and network-attached storage. It’s hard to know where SDS starts and ends.

Like other software-defined computing, SDS is heavily dependent on abstraction. Hiding the hardware under a layer of virtualisation is essential to SDS. Abstraction is what everyone agrees on.

The dispute comes in terms of defining what type of hardware should lie beneath the hypervisor.

According to Stephen Parker, head of cloud strategy at local aggregator Rhipe (formerly NewLease), the SDS world is generally divided into two sectors: the “closed ecosystems” that prefer specific hardware under the hypervisor, and the “open systems” that don’t favour specific brands.

The closed systems are dominated by the well-known brands such as Hewlett-Packard and EMC, while open systems include open-source vendors like Red Hat and proprietary but software-only players such as DataCore and VMware.

Parker warns that while many of the closed systems are tolerant of a variety of hardware, it’s an inevitability that the vendor’s own kit would run best on its software, which is not a criticism, just a natural consequence of knowing your own product better than others.

Parker points out that one approach is no less valid than the other.

“You have to ask what situation a reseller or a customer has currently,” says Parker. “If an environment already has a strong relationship and investment with a particular vendor, then it makes sense to continue with them and get the continuing support.”

The ‘purists’ of software-defined anything would say that the ultimate aim is to be hardware-agnostic – to the point of running commodity hardware in a virtualised pool. To this ideal, DataCore Software’s chief executive George Teixeira predicts that this year, servers will replace traditional storage arrays, with the help of SDS such as his company’s Virtual SAN product.

“Servers will combine with software-defined storage and continue to build momentum for a new class of ‘storage servers’ and hyper-converged virtual SANs,” says Teixeira. “The latest generation of servers is powerful and will continue to support large amounts of storage.

“Virtual SAN software is maturing rapidly and it will further drive the transformation of servers into powerful storage systems that go beyond today into full-blown enterprise-class virtual SANs.”

It’s not a surprising opinion from a software-only vendor, but others are sceptical about the role of commodity hardware in the space.

Distribution Central’s technical manager, Richard Denyer, reminds us there are kits from which you can build your own car. But he says very few people build their own vehicle for obvious time and reliability reasons. “You shell out for an already-assembled car because you know it works and you have support if anything goes wrong.”

In the same way, Denyer questions the value of resellers and customers spending time and effort to convert commodity hardware into storage.

“Data has gravity. Data is important. Do you want to go through the hassle of working out what hardware goes with what software? Even with the latest release of a software-defined storage product – and I’ll pick on VMware’s vSAN as an example – not every piece of hardware works well with vSAN,” says Denyer.

“The internet [highlights] people complaining about how ‘hardware X’ doesn’t work with vSAN. As a customer do you want to go through that? You can buy a server and purchase whatever SDS product and go through the process of making it all work. But what happens when something goes wrong? Who do you ring?

“Storage is important. Commodity storage is the same as kit cars – not everyone’s a mechanic,” says Denyer.

Melbourne systems integrator Perfekt deals with all the major storage vendors. Senior integration specialist Cristian Danci agrees the channel should focus on abstraction when thinking about software-defined storage, and get their minds off commodity hardware.

“Even if you use commodity hardware, the software is still proprietary. That’s a critical dist-inction [to make],” says Danci.

“We’ve found a whole bunch of [storage] customers where they’ve said, ‘Just give us the software and we’ll go and build the hardware’. But then a lot of them come back and say, ‘The hardware is too hard to do, please give us the hardware’.”

“Even though it’s just commodity hardware, the engineering teams still need to determine what processor should be used, what RAM should be used, what interoperability it needs, how it will best perform under specific configurations...That’s really, really hard to do.

“That’s why the vendor’s own storage hardware still works the best, because their engineering teams have done all the work in actually getting the hardware to work, which is a really hard process to do.”

Danci also points out that the commodity market moves very rapidly, and it’s difficult for even the most diligent SDS vendor to keep up with making their software fit the nuances of every new box.

When the industry was first turned onto the potential of SDS, the role of commodity hardware was poorly understood, says Adrian De Luca, chief technology officer of Hitachi Data Systems Asia-Pacific.

“A couple of years ago there was a misconception of reduced cost resulting from using commodity hardware with software-defined networking,” he says.

“Now it’s all about agility – the ability to scale out and call for additional services at will.”

In defence of commodity hardware’s potential in SDS, Westcon’s manager of innovation and services, Darryl Grauman, says there are “self-healing” storage utilities now emerging. Such tools will make commodity hardware more reliable, but he warns that until this nascent field of technology matures it is best to stick with specialist-storage hardware underneath a software-defined storage system.

Next: Why bother with SDS?

So if you can’t save a few dollars by mixing in white boxes for storage, what is software-defined net-working good for? You might as well just buy a traditional storage area network, right?

Darren Ashley, director of CRN Fast50 reseller BEarena, presents some compelling arguments.

“With a traditional SAN, the performance peaks on day one and it’s all downhill from there,” he says. “Also, the cost of scaling up is terrible. Say you wanted 50TB for the year. But for cost reasons you bought 30TB initially then added 20TB later – well, you end up paying a lot more doing that than just buying 50TB from day one. We’ve always seen that as a frustration.”

For BEarena, the solution is hyper-converged appliances from Nutanix. Such technology provokes fierce debate as to whether it can be considered software-defined storage. But the scalable boxes work for Ashley and his team, who have had tremendous success as an early Nutanix adopter, with clients such as Hyundai to their credit.

Another reason to consider SDS is the rise of mixed media. Flash storage is now well and truly in the mainstream, but the additional cost still means many resellers and end customers would need to have both in their IT environments. With SDS, controls can be put in place to direct certain workloads to flash and other I/Os to spinning disk.

“The [flash] industry would have us believe that customers will shift 100 percent to all flash, but it is not practical due to the costs involved, and the large installed base of storage that must be addressed,” says DataCore’s Teixeira.

“We will need smart software that has the feature stack that can optimise the cost and performance trade-offs and migrate workloads to the right resources needed, whether flash or disk. Software-defined storage done right can help unify the new world of flash with the existing and still-evolving world of disks. Both [of these] have a future.”

Hybrid cloud also presents fertile ground for SDS.

VMware is an example of a vendor that has adapted to this trend. In the past, the company actively discouraged its channel from using Amazon Web Services. But in November, VMware was seen on the exhibition floor of AWS’ re:Invent conference spruiking how its vRealize Suite cloud management product seamlessly deals with the Amazon cloud as well, as its own vCloud Air.

“What has changed during that time? Essentially, VMware has canned the vitriol and decided to position itself as a ‘true’ hybrid cloud provider, according to partners,” reported CRN US at the time.

If VMware can see the light, it won’t be long before your customers do.

It’s worth mentioning that there are parts of the industry that don’t yet “believe” in SDS. US vendor Quantum is an example, with senior vice-president of strategy Janae Stow Lee saying the confusing variations of SDS from different vendors make it more difficult for customers and partners, not easier.

“At a time of increased demands on IT staff and continued budget constraints, it’s unlikely that customers will have or be able to maintain all the skills and expertise to understand – and trade off – the wide variety of IOPS, throughput, latency, cost and durability features different options provide,” she said. “Sometimes flexibility just equals complexity – and that means many software-defined solutions won’t make the cut because they just plain aren’t easy enough to understand.”

For those resellers that do want to experience SDS, some of the major vendors – NetApp (Clustered Data ONTAP) and EMC (VIPR), to name two – are now including access to their software with new hardware sales. Whether the software layer remains as a conventional “single node” controller or if a true SDS is implemented is up to the partner.

Westcon’s Grauman tells us: “It’s time for everyone to dip their toes in the water and give software-defined storage a try. Next time you put in new storage, just put in a virtual layer on top.” 

FACTFILE: Hyper-converged

There is considerable debate as to whether hyper-converged boxes such as those from Nutanix and Simplivity can be considered software-defined networking.

Hyper-converged appliances combine storage and compute, with those resources from all the boxes combined by a software layer to act as one large pool. Scaling up or down is a simple matter of plugging more boxes in or taking boxes out.

As a prominent Nutanix reseller, BEarena absolutely considers it a viable storage solution.

“Different models of Nutanix have different ratios of compute and storage,” says director Darren Ashley. “So if your focus is on storage, you get boxes that are geared towards that, and then you have software-defined storage.”

Westcon’s manager of innovation and services, Darryl Grauman, says that hyper-converged devices are excellent in certain situations but can be costly for to implement inside of existing environments.

“They’re not cheap. You pay for compute, storage and integration of the two with hyper-converged devices. So they’re convenient for new ‘greenfield’ set-ups,” Grauman says. “But for a client with existing infrastructure who wants to fill a specific gap, you might end up paying for things that are redundant.”

Storage accelerators such as PernixData are also looking in from the fringes of software-defined storage. PernixData provides a software layer that intelligently directs I/O onto different types of storage to optimise performance. The vendor’s claim to fame is that it can use server RAM as storage – reminiscent of the 1990s IBM-compatible PCs running virtual disk – for lightning-fast performance for apps that need it.

FACTFILE: Storage

CRN’s sister publication, iTnews, has produced The Buyer’s Guide to Primary Storage, an excellent read for those who want a crash course in the world of storage (you can download it at bit.ly/1yxFwzE). Courtesy of the guide, here is a quick run-down of the different categories of storage:

Traditional: The majority of customers buy traditional types of storage. These systems contain Redundant Arrays of Inexpensive Disks (RAID) of mostly spinning disks. They tend to be 7.2k, 10k, and 15k RPM disks, connected with SAS, SATA, or fibre channel connectors for some of the older-style arrays.

Flash: Provides a step-change in low-latency performance. Spinning disk latency is limited by the physics of moving parts: the fastest 15k RPM disk will only give you about 4ms average latency; 2ms for both seek time and rotation delay. Flash, meanwhile, can regularly and consistently achieve well under 1ms latency. There are no moving parts, so you’re limited by a completely different sort of physics, at the electrical and, yes, even atomic level.

Hybrid: An attempt, often successful, to achieve “the best of all possible worlds” in a single device. These products work in unison to combine the capacity and price of spinning media, which is then supplemented by Flash and memory to improve the performance.

Server-SAN: is a term coined by Wikibon to describe the recent practice of putting physical storage inside servers (as was the common place before dedicated storage devices) so that it is closer to the CPU, and aggregating the storage across servers using software.