Joshua Resnick - Fotolia

Net gain? Why networks matter when it comes to hyper-convergence

The rise of hyper-convergence is radically changing how enterprises manage their IT estates, but its effect on the network has yet to be fully realised

With the rise and rise of hyper-converged infrastructure (HCI) in many enterprise contexts, is network management getting much easier? Not always, not yet – but there’s plenty to explore in the idea.

HCI, remember, has gained profile – and market share – as an out-of-the-box proposition that marries virtual servers, storage and networking to provide a seamless view of the enterprise IT stack. But is that marriage really as convenient as it sounds?

Is it changing the world of network management for the good, as well as data systems architecture – or is it more a question of horses for courses? Sometimes it’s a fit and sometimes it isn’t.

Efficiency and HCI

For Chris Gardner, an infrastructure and operations senior analyst at Forrester, the first point to make is that HCI’s impact on any IT stack and network will vary enormously.

“In an efficient organisation, the network team works in tandem with storage and compute before a roll-out. The network managers must understand that this is not a typical server – it will have higher bandwidth, resiliency and management requirements. That means work needs to go into ensuring the configuration of the network within the HCI fits with enterprise standards and policies.”

And what’s the best way to do this? With software-defined networking (SDN), says Gardner.

“The network team will also typically hand some of the network responsibilities to the owner of the box. Network engineers don’t really want to be in the business of managing HCI network fabrics every day – just as, similarly, they weren’t often involved with per-VM network management.”

But if those are some of the likelier practices around HCI and network management, what happens in the real world?

“Where things get challenging is that most organisations are not efficient ones,” says Gardner. “There are plenty that treat networks as a siloed discipline. The network team may not, for example, even talk to the compute and storage teams that often. Where that’s the case, however, there may well soon be trouble – whether in the form of oversubscribed capacity or HCI network fabrics configured well outside of standard network policy.”

Time is your friend – plan ahead

Not surprisingly, the first thing Gardner and other experts recommend to those taking a serious look at HCI is to get the network team in the room early.

“They need to be involved as much as possible with the planning and implementation. This can't be sprung up on the networks team on the day of deployment,” he says.

“The network in an HCI is like a microcosm of your enterprise network. If you block traffic on certain ports enterprise-wide, you need to abide by those rules on the HCI. If you lock things down a certain way on the enterprise network, this should be done in a similar way.” Quality-of-service settings on the network shouldn’t conflict with those on the HCI, either, says Gardner.

If this seems obvious to you, great, but many consultants make the point that far too often implementations see HCI behaving differently to the rest of the network – and then everyone scrambles to figure out why. It’s here that SDN can help by applying consistent policies across multiple types of hardware.

A living project

Another thing to say about HCI roll-outs and the network is that they require regular attention. In other words, a network manager doesn’t typically set up an HCI and forget it. You’ll be tweaking settings well into the future – just as you would an enterprise network. Again, some of these responsibilities may eventually filter down to the traditional compute and storage admins, but that needs to be done in a controlled way.

“What we’ve seen work overall is having more generalist admins,” says Eckhardt Fischer, a senior research analyst at IDC. “They could be network experts but also have just enough knowledge to troubleshoot compute and storage. They might be storage admins but know a little about network in compute. This works far better than having specialists stuck in silos.”

Keep it simple, stupid

A further point to make about any HCI roll-out is that there will be benefit in adopting a platform that is similar to existing infrastructure. Yes, a completely different interface on the HCI may look cool in the demo, but it can be a nightmare to manage, not to mention having to deal with additional supplier relationships.

One way to address this is to consider staying with a current supplier. You’ve Cisco network hardware? Consider Cisco for HCI.

In practice, many will naturally take this path, says Barry Coombs, operations director and virtualisation specialist at the IT services provider ComputerWorld.

“The appeal of HCI for most IT buyers is how quickly it works out of the box, promising a simpler life in maintenance terms, at least eventually, with easier scalability. That’s the lure with HCI in many respects – how it is pre-packaged by the manufacturer, with easy upgrades and the adding of nodes.”

But, Coombs adds, most likely it’s an option that will need a 10Gb datacentre rather than 1Gb, even for small organisations.

“The idea in many respects is to simplify the network, for example by moving from a blade platform while expanding the software-defined datacentre. But it’s clear there are lots of variables for the networks manager to consider with any migration from physical to virtualised workloads, and any shift towards software-defined datacentres. Micro-segmentation also comes into this picture, with NSX one clear option for delivering this.”

Embracing HCI

In many ways, says Coombes, the embrace of HCI goes hand in hand with the rise of more generalist IT support internally in many organisations, backed by supplier support.

“That’s what many organisations will be planning to do in adopting HCI. For example, in the public sector, where funding has been shrinking, the advantage of HCI is that it should be easier to manage with a simplified support team. That’s the end-goal for many.”

But while Coombs is a fan, he emphasises it’s not a one-horse race when it comes to choosing.

“There are simple SAN solutions that need minimal specialist knowledge and offer good performance. In fact, at ComputerWorld we have even replaced some early HCI and installed flash-based SAN in its place.”

HCI at the University of Reading

What’s the story when HCI is adopted? At the University of Reading, the research computing operations are overseen by academic computing team manager Ryan Kennedy, who heads a six-strong team.

In 2017, Kennedy’s team migrated academic support workloads from a mixed collection of legacy platforms to a brand-new “research cloud” built on a single cluster of Dell EMC XC Series appliances running the Nutanix Enterprise Cloud Platform, and using Nutanix AHV Hypervisor and Nutanix Self Service Portal.

“What drove the move was the need to consolidate and simplify our infrastructure,” says Kennedy. “The team was supporting a mix of five different compute and 10 storage platforms, and it took a lot of staff hours just to maintain them all.”

One option Kennedy looked at was to move everything into the cloud, but the subsequent analysis revealed that was very expensive, so Kennedy opted instead for Dell EMC XC Series appliances running the Nutanix Enterprise Cloud Platform.

“It has enabled us to deliver the equivalent of a public cloud service – like AWS – right in our own datacentre.”

Several hyper-converged platforms were considered for the project, with the Nutanix platform chosen, says Kennedy, for its linear scalability and virtualisation capabilities – all with integrated management through a single interface. The ability to automate management processes and make compute, storage and network resources available directly to users via a self-service portal was also crucial.

“Once academics have secured funding for a project, we allocate resources on the Nutanix Enterprise Cloud Platform, leaving users free to configure and manage their virtual machines, storage and network connections via the portal. It’s a good outcome because it frees up staff to support academic users with the design, running and support of those workloads, rather than spending all their time keeping the infrastructure lights on.”

Read more about HCI

  • HCI and the cloud have virtualisation at their cores, which makes the two platforms a good combination for organisations planning a hybrid cloud IT architecture.
  • Implementing HCI systems in the datacentre is a big undertaking for IT. Learn the basics of this emerging technology and its components before you make the leap.
  • HCI's easy implementation and management help admins solve some of the challenges of edge computing, such as distributed endpoint management and a trend toward data silos.

As part of the migration, Kennedy and his team switched from using VMware to the integrated AHV hypervisor. This further simplifies virtualisation deployment, with the Nutanix AHV Hypervisor and virtual machine resources all managed from a single Nutanix Prism unified management plane. It also removes the need for any VMware software licensing, maintenance and support.

So what were the network issues Kennedy needed to work through – or was addressing – along the way?

“We have several compute clusters and they were struggling to deliver. The network sometimes could not cope with the volume of traffic from the SAN to the compute clusters – and the way the clusters were built made changing them a hard step to take.”

Kennedy points out, too, that his team is just an element of a central IT organisation at the university.

“What’s good about the research unit is that you get a year’s head start to adopt newer technologies. We can experiment as we go along, to an extent. With this roll-out, we migrated to new Dell EMC networking infrastructure at the same time, as part of a big capital investment, including new switches and routers. But we also built the cluster to be separate from the main corporate infrastructure, so it’s well contained – and easier to configure and build.”

Fast growth

It’s growing quickly, too, moving from five nodes initially to 18 very quickly and a step-up to 25 nodes planned. “And have two 10Gb Ethernets and the Nutanix HCI load-balances the traffic. It is a lot more robust that our previous set-up.”

The on-premise HCI will deliver for at least the next three to five years, says Kennedy, by which time he thinks a scale-up to public cloud should be viable.

“The ultimate future for us is cloud, but on-premise is cheaper for now – and we know we can scale out the cluster using Nutanix HCI. It’s just the right setup, and we only need two full-time equivalents to maintain the data and ensure resilience.”

HCI and the future

If that’s the future for the University of Reading’s research cluster, what about HCI itself?

Forrester’s Gardner says this: “HCI has been growing and will continue to grow. A couple of factors are in play. The HCI model better fits the multi-tenancy model that many enterprises are now deploying, while performance is on par, if not better, than piecing together traditional stacks. In short, it’s a more efficient use of space.

“Probably the biggest indicator, though, is that more organisations have moved or are moving business-critical apps to HCI. Once a few enterprises in a sector adopt, everyone tends to follow. If networks are your thing, you should look to be prepared and involved at every step.”

Read more on Data centre networking

Data Center
Data Management