Converged infrastructure products, integrated computing platforms, single-stack solutions, data centre in a box: whatever you call them, the benefit of buying storage, servers and networking in a single, pre-engineered bundle is that, vendor-wise, when something goes wrong, you've only got one butt to kick.
That's not quite how IT vendors put it, of course: they're more likely to talk up the benefits of time-to-deployment or single-pane-of-glass management of deploying products such as Vblock from the VCE coalition of VMware, Cisco and EMC, or IBM PureSystems or NetApp Flexpod or HP CloudMatrix. At a push, they'll mention the advantages of having a "single throat to choke".
More ReadingFailover in FOUR SECONDS? HP's SAP-specific iron hits the streetsOnly the Chinese can save us now, trill server vendorsSpace-saving ideas for improving storage performanceSimplivity shows off VDI configs with Nvidia... Doh! VMware, what are you doing here?Facebook adds Flash to up the tempo of its enormous disk-o-tech
But however the benefits are described, storage has taken a back seat in customer decisions to invest in these products, which accounted for just under 4 per cent of data centre infrastructure spend in 2012, according to IT market research company IDC.
Servers, by contrast, have been in the driving seat as IT teams focused their efforts on consolidating the workloads previously performed by many smaller machines through virtualisation and migrating them on to fewer, larger hosts. That, however, is starting to change as legacy storage infrastructures run into problems.
First, the wholesale virtualisation of servers in data centre environments has made previous approaches to storage management unsustainable. In some cases, the savings organisations make by consolidating servers end up being used on extra storage capacity, as default sizes for virtual machine disk images get set unnecessarily high by over-cautious admins. For many, that has resulted in vast over-provisioning and under utilisation of networked storage.
Then, there are the accompanying performance problems to contend with. If you stick 20 VMs on a physical host, their I/O streams start to compete. Traditional networked storage falters as spinning disks struggle to respond quickly to large amounts of random I/O.
Second, there’s the ‘big data’ problem: most companies are simply trying to store and manage more data than they have ever before, but most recognise that simply buying more will not solve the problem.
In other words, legacy storage architectures are costly, cumbersome and complex to manage. Plus, they’re increasingly placing a roadblock in front of further server virtualisation - and that stands in the way of further data-centre consolidation and the migration of new workloads to the cloud.
Storage: the 'next big bug-bear'
“The focus for our clients is now very much on storage,” confirms Alex Mcloughlin, a director at systems integration firm Union Solutions. “A lot’s been done already in server consolidation and many organisations feel there’s not much more to be gained there right now. But storage is the next big bug-bear to tackle and they’re looking to abstract it by building a software layer that will enable them to make storage decisions based on cost per gigabyte and the service-level requirements of individual applications,” he adds.
It seems they’re looking for an approach to storage that eliminates traditional silos. Converged infrastructure products that combine compute and storage resources on the same physical infrastructure, with common management capabilities, answer that need. As a result, storage is becoming a more important consideration in converged infrastructure buying decisions and may even replace servers as a driver of platform choice for some IT departments, says Mclouglin.
Many of the reasons IT teams have already invested in converged infrastructure stacks are as applicable to storage as they are to servers. For a start, there’s that ‘one arse to kick’ argument: with a CI stack, there’s one number to call and one support organisation to speak to when something goes wrong. The finger-pointing that often goes on between storage, server and networking vendors unwilling to ‘own’ a particular problem is avoided. That can be a powerful draw, says Mike Price, a virtualisation specialist at systems integration firm Logicalis.
There are other advantages, too. Converged infrastructure products are much quicker to deploy because they arrive pre-integrated. “Whether IT likes it or not, my perspective is that people in the business want systems up and running quickly,” says Valdis Filks, an analyst at Gartner.
“The question for them isn’t whether the IT team is capable of ordering servers, storage and networking equipment separately, taking delivery, putting it all together and testing it, but just how long that process is likely to take,” he says. “Implementing standardised applications often doesn’t justify that effort. Bosses would rather IT passed the screwdriver to the vendor.”
At this point, it’s worth pointing out that there are various subtleties in terms of the extent to which systems are pre-integrated and pre-configured. Analysts at Gartner, for example, make fine distinctions in the converged infrastructure category between Integrated Infrastructure Systems, Integrated Workload Systems and Integrated Reference Architectures. The differences lie predominantly in whether the kit is assembled by the OEM or further down the chain by a VAR, and whether or not applications are included in the original bundle or added later by the reseller. From the point of view of most end-customers, however, the experience is roughly the same: unless it’s a very large shop piecing a Reference Architecture together, it gets a package from suppliers that is ready to run.
While CI stacks are rarely cheaper in terms of upfront capital investment, they may offer reduced cost of ownership over the long term. Pre-validated configurations mean fewer interoperability issues. Patches are only released to customers once the vendor is sure that patches for one part of the stack won’t create problems with other parts of the stack. And the fact that all that technology is consolidated into one ‘tower’ - a dense, self-contained enclosure - typically means a smaller footprint in the data centre and lower costs for power and cooling.
Finally, CI stacks come with a single management console that puts all the tools needed to run the disparate parts of the infrastructure in one place. That, according to vendors, saves time and lowers complexity for admins. In fact, executives at HP claim that it will ultimately render titles such as ‘systems admin’ and ‘storage admin’ redundant as they are replaced with ‘converged infrastructure admins’ - a potentially ominous prediction for some.
Then there are the advantages specific to the storage environment when it becomes part of a compute/storage/networking stack.
Converged infrastructure stacks deliver an elastic pool of storage resource that’s relatively easy to expand with minimal overhead and into which it’s possible to incorporate higher density storage mediums, such as Flash. A single enclosure can typically house multiple controller nodes and multiple enclosures can be clustered for backup and disaster recovery.
The software automation layer of the stack, meanwhile, typically enables it to handle the performance demands of file-level storage and the scalability demands of block-level storage.
Above all, there’s an important performance benefit to be gained over previous networked storage approaches because traffic remains within that enclosure, transferred over a shared backplane for lower latency and great speed.
“In part it’s a simple question of physical proximity: bringing storage disks closer to applications and to CPU and to reduce the redundant layers that exist between them,” says Robin Kuepers, storage director for Dell in EMEA. “A different internal switching architecture, converged on the backplane of a smaller form factor opens up the possibility of higher bandwidth and throughput.”
What is more, available storage capacity is used more efficiently, claims David Leyland, head of architecture, data centre and cloud at Dimension Data. “Even the smartest and most competent of IT teams struggle with storage management, not least in having the visibility into where everything is stored and where available capacity lies,” he says. “With a converged stack, all your available capacity is in one place and is visible through a single pane of glass console, so there’s inherent simplicity there. In our experience, it’s pretty common to achieve storage utilisation rates northwards of 70 per cent or 75 per cent with these products.” That compares with an average of around 20 per cent with networked storage arrays scattered around the infrastructure, he adds.
There’s also a benefit in terms of storage administration tasks, says Chris Johnson, EMEA vice president and general manager of HP’s storage division.
“I talk to a lot of customers and they just don’t want people in the IT department analysing levels of firmware between storage and servers and HBAs [host bus adaptors]. These are fairly low value-add activities. They want IT teams focusing less on engineering and more on what the business needs,” he says.
A new approach to convergence
While most of the large, established systems vendors have introduced converged stacks in recent years - most recently Dell with the VRTX tower for small and medium-sized business in July - a new breed of suppliers has also emerged that take a more storage-centric approach. These include Nutanix, Simplivity, Tintri and Virsto.
“Storage-centric design makes a fundamental assumption [that] it is easier to insert standardised servers into an optimised storage environment than it is to build a really good VM-optimised storage environment,” says Forrester analyst Richard Fichera.
By integrating server resources more tightly to a VM-optimised storage resource, these new products, he believes, “have the potential to radically change the balance of power in the enterprise infrastructure market by allowing vendors with strong storage offerings to rapidly encroach on the previously well-defended markets of established server vendors”.
In the case of Nutanix, its basic module is a 2U enclosure containing four independent 2-socket servers. Each module contains four system nodes, each with a 2-socket x86 node with 300 gigabytes (GB) of flash cache and 5 terabytes (TB) of user-visible disk on each node, giving a total of 20TB. All data communications between nodes, as well as between additional nodes that can be federated onto a single logical system, use a dedicated 10 GbE switch.
“This is truly software-defined storage,” says Dheeraj Pandey, CEO of Nutanix. “All our intellectual property is in software. We take simple hardware components with that software on top to make it just as good as any enterprise-grade infrastructure product out there with the benefit of getting rid of special-purpose storage appliances.”
Nutanix and its immediate competitors are largely early-stage companies and, for many enterprise IT buyers they represent a step into the unknown. But what they’re proposing shows such promise for reducing many of the barriers to effective implementation of a virtualised infrastructure and private-cloud computing, says Forrester’s Fichera, “that the rest of the industry will almost certainly be in hot pursuit”.
One overriding question remains, however, for those in charge of data centre infrastructure investments: to what extent does a converged stack from a single vendor, regardless of whether it takes a server-centric or storage-centric view, limit our future choices? In other words, what are the risks of vendor lock-in?
For many, those risks are a real worry - but maybe they shouldn’t be, according to Leyland at Dimension Data. “I’m not convinced much changes. Reduction of complexity always carries a certain amount of lock-in, but not necessarily a more than before. If you historically ran an EMC infrastructure for storage and you wanted to transition to HDS, it was pretty much a Grand Slam to make that happen. At least with blocks of converged infrastructure, you’re able to migrate block by block,” he says.
It’s a view echoed by Valdis Filks at Gartner. “To be honest, if you go out and buy a storage array from one vendor, you’re locked in for the next four to five years, because realistically, you can’t really get rid of it until you’ve depreciated it anyway,” he says.
Converged infrastructure buys represent a relatively minor element of overall data-centre investment, albeit a growing one, and most companies are still willing to buy and integrate ‘best of breed’ components. But vendor lock-in, he says, is to some extent an inherent risk of any IT procurement activity. “At least with a converged stack, you’re passing the screwdriver to the vendor,” he says. ®