This article is more than 1 year old
Is hyperconvergence about to take over the enterprise data centre?
After the headlong rush into HCI, where next?
Limits on hyper-convergence scale-out
What limits are there on hyper-converged scale-out? Abbott sees HCI networking as one. "Networking is mostly ignored by almost all of the HCI providers,” he says. “Most don’t include it in their offerings and simply presume that 10 Gbit Ethernet will meet customer needs. In small deployments, this is a safe assumption, but as volumes of data grow, the networking connection may become the bottleneck in maintaining performance."
"Different vendors have strategies for ‘intelligent placement’ of workloads to minimize the off-box performance hit, but they may all reach limits when replication and application traffic compete with basic storage I/O."
Locke relates scale-out limits to application performance consistency. "Applications, and users, want performance consistency. It really all comes down to a question of at which scale does service inconsistency become an issue? This varies greatly depending on how metadata is managed and storage access controlled across nodes."
Sakac has his own take on scale-out. "Scale is not a factor – the very nature of hyper-convergence allows it to scale at different magnitudes. But as you scale you have increasing requirements on a hyper-converged solution that determine the choice of underlying components. Networking is the backbone of hyper-converged systems and becomes critically import at data centre scale and needs to be part of the hyper-converged management domain,” he says.
"And, to deliver full-service IaaS for all workloads at enterprise scale, a hyper-converged solution needs to additionally include software lifecycle management and application resource allocation management. These requirements don’t change the applicability of hyper-converged, just the boundaries of the hyper-converged system at scale."
Unsuitable use-cases
What, if any, applications and use cases are ill-suited to hyper-convergence?
Here's Locke on hyper-convergence use-cases: "Until recently most vendors targeted HCI solutions at a small number of use cases, most notably desktop virtualisation solutions and ‘VDI’, general server virtualisation and ROBO, remote office/branch office. Now a small number of vendors are beginning to position some HCI systems as suitable for more generic enterprise workloads, but does require very good/consistent service quality and granular service quality control."
Logen says, "As with traditional virtualisation, not everything can or should be virtualised. Low latency or multi-cast network dependencies, non-x86 workloads or extreme storage IO-generating applications are still more suited to non-converged (dedicated) hardware environments."
Some legacy IT apps are contra-indicated in Sakacv's view. "Today hyper-convergence solves the plurality of enterprise requirements, but there are some workloads with very specific requirements that that fall outside its current capabilities. These are the most classic, legacy IT applications written over the last 15 years that depend on specific infrastructure services. There is no software-defined storage that addresses this today the way that traditional converged infrastructure does."
Hyper-converged data centres
Will data centres go all-in on hyper-convergence or have a mix of non-converged, converged and hyper-converged systems?
"In my long IT life I have never yet seen any IT solution that is perfect for every use case,” says Locke. “HCI will grow the workloads for which it is a good fit as solutions mature, but I cannot see it ever being perfect for all. There will always be ‘edge cases’ where doing something specific to the workload will justify running a separate platform."
Logen sees HCI limits in the data centre too. "It's likely converged will go into decline due to the inherent benefits of hyper-converged. However, there will always be a case for traditional non-converged infrastructure for applications that require dedicated resources."
Abbott doesn't see data centres adopting wall-to-wall HCI: "Over the shorter term, we expect adoption to be chiefly concentrated in small to midsized businesses, while large enterprise adoption will be mostly focused on ROBO and departmental-level deployments. Even if this is as far as HCI ever gets, this is a substantial market opportunity amounting to several billion dollars.
"The question of whether HCI can make a significant dent in core enterprise data centre environments remains much more uncertain, however. There is little doubt that this is going to be a harder sell overall. Large enterprises generally have more sophisticated requirements across a broader range of applications – many of them homegrown – and a ‘one size fits all’ infrastructure is a more difficult sell, especially if many of those core applications have still to be virtualised."
Sakac is like-minded, saying, "CIOs will use different architectures to meet the cost, flexibility and reliability requirements of diverse workloads. Hyper-converged gives one architecture likely to end up meeting the bulk of requirements, but will not meet all and CIO’s will have need converged infrastructure systems to address these other requirements."
Technology development
How will hyper-convergence technology and usage develop? Logen thinks HCI will become mainstream. "As more vendors enter the market or evolve their existing converged solutions into hyper-converged offerings, hyper-convergence will become the new normal. Just as converged did five plus years ago."
Sakac sees HCI potentially overtaking converged infrastructure. "Hyper-converged will continue to grow exponentially and is growing faster then the converged infrastructure market which is currently four times larger then that of hyper-converged. The need to deliver an easily supportable, consumer experience is expanding to encompass more and more use cases. And technology changes will continue to deliver better economics as well as some of the advanced capabilities not available in hyper-converged today."
Abbott views HCI networking development as crucial: "So far it’s been ignored by most HCI vendors, though it’s a tick-box requirement for full-blown integrated infrastructure systems. Until the east-west traffic problem and integration of software-defined networking frameworks like VMware’s NSX are addressed more seriously, the perception that HCI is best suited for small business and mid-range customers, or branch deployments within larger companies, will likely persist.
And beyond HCI? "The so-called 'software-defined data centre' promises to bring together virtualized compute, storage and networking across the entire data centre, including both on-premises and cloud-based infrastructure. We are a long way away from that."
Locke's dry comment on how he expects HCI technology to develop? "Rapidly." ®