Converged infrastructure is now a mainstream proposition for enterprise class companies and most of the major vendors have been pouring resources into the technology for some time. Meanwhile, there is also growing interest in hyper-convergence with many companies wondering which will suit them best?
The answer depends on a number of factors, scale, ambition and budget not least amongst them. But according to Francis O’Haire, director for technology and strategy with Data Solutions, first it is important to make a distinction between the two ways of organising infrastructure.
Lumped together
“They tend to be lumped together but they’re actually very different. Converged infrastructure has been around for a long time in the form of VBlock, FlexPod etc, where vendors marry together various products within storage, networking and compute to put them all into a rack, put a single view on it and maybe a single support contract. That gets boxed up and shipped off,” he said.
“In reality, inside that rack are all the pieces that you would traditionally find inside a legacy data centre—you still have storage area networks, the networking to go with that and servers or blades. It’s actually no different than just putting it together yourself, other than the fact that you have one support contract to look after it all.”
In reality, converged infrastructure organised this way has the same limitations that standard three tier architecture has, in that eventually the storage area network (SAN) component is going to run out of power while the compute aspect remains easy to scale.
HCI
Hyperconverged infrastructure, by comparison, completely changes that model and features architecture that is organised in a way that is more similar to the way Google, Microsoft and Amazon build out their data centres.
“In a hyperconverged model there are no SANs whatsoever, the infrastructure is built up of standard commodity x86 server components and each one of those runs with intelligent software on top that contains analytics capabilities and scaling but also storage technologies that allow you to make use of all the local disk and flash within the server nodes. It allows you to spread that storage fabric across all of the nodes without using any SAN techniques.”
The appeal of moving to hyper converged infrastructure is that companies want to reduce the complexity of their own data centres.
“If you build a data centre using storage fabrics, SANs and a compute layer on top, then you’re using an architecture that has been inherited from the days of the mainframe. The problem is that things have moved on. For example, virtualisation is now almost ubiquitous and it stresses that architecture to its limits,” said O’Haire.
“Virtualisation concentrates a lot of compute resources in a small amount of space and all of those resources, usually virtual machines, all want to talk to the SAN. Unfortunately, in converged infrastructure, the SAN doesn’t scale as easily as the compute layer. If you want to scale the compute layer, you just need to add another blade or another server and you can keep going with that, but at some point, you’ll outgrow what the SAN can provide to the compute layer.”
Big now
Hyper-convergence is becoming very big in the Irish market for a number of reasons, according to Catherine Doyle, enterprise director of Dell EMC Ireland. The first is that it is really quick to deploy—it can be plugged in and ready to go within 15 or 20 minutes of arriving on site.
“There’s no configuration, no putting things together and you can start small and expand which means you don’t have to do a huge sizing exercise for your applications to figure out what will we need in three years’ time,” said Doyle.
The ability to start small and scale as you go, plugging in modules as required is “hugely attractive” to companies looking to get up and running very quickly.
“There’re various levels of it, so from our point of view we have the VX Rail which will scale to a certain amount of virtual machines. It’s built on VMware technology so it’s built on VSAM so hyper-converged is the software-defined ‘message in a box’, if you like,” said Doyle.
“Rather than taking the software to create your own structure, you buy an entire solution as a product. For workloads that you cannot put in the cloud, it gives you a cloud-like experience for a fraction of the cost and it doesn’t take a lot of technical people’s time putting it together, figuring out how to get it to work in the first place.”
Money saving
In cash terms, Doyle explained that it can save companies money through streamlining the entire provisioning process.
“Another big factor is the whole new DevOps environment for companies as they move to next-generation applications. They need rack-scale type structures and they need to create environments very, very quickly from a DevOps point of view. We’re finding that really taking off now,” she said.
While the tech industry has been talking about DevOps for some time, Doyle suggested that it is only really beginning to become a reality now.
“We’re now seeing it with customers where they are actually building next generation applications and traditional IT provisioning, like command-line type provisioning, is just not fast enough,” she said.
“They need to be able to automatically provision it by the slice. So rather than trying to build it and then figure out how to put something on top of it, to be able to provision it by the slice makes it much easier. That makes it so much quicker to get an application up and running.”
Logically built
Meanwhile Matt Foley, director for EMEA solutions and technology group for Hewlett Packard Enterprise, suggests that part of the appeal of hyper convergence is that it builds logically upon conventional converged infrastructure.
“If you go back a few years, the most popular enterprise application was VMware and the way to run VMware was typically on blade servers connected to a SAN normally with fibre channel,” he said.
“And in order to do that, you needed people who understood VMware, people who understood servers, people who understood networking and people who understood SANs—so that’s four people and you could justify that when you had one application per server.”
However, once IT departments started putting 20 or 30 applications on each server, things become more complicated.
“If I want to spin up one virtual machine, why do I have to talk to four people? That challenge of adding capacity or changing settings really led people to start innovating in that area,” said Foley.
“That was the beginning of a lot of ideas in this area all geared towards making it easier to manage and maintain this environment without having to have a committee meeting every time you want to do something new.”
And one of the other challenges was site planning.
Site planning
“So, you have all this equipment and you have to co-ordinate it across four different departments and for example if the SAN guys are late, then the whole project is late and the potential for problems is greater.”
According to Foley, part of the appeal of hyperconverged infrastructure is that it can be purchased preconfigured and ready to go.
“That’s where our VCE technology really came into the market—they were pre-tested and pre-qualified hardware bundles with a defined specification in block format. So, you could then say ‘I need this many virtual machines’ and that would mean you needed four V-Blocks,” he said.
“You could just order them and they came complete with storage and networking switches and servers and they were all tested together. That is important because one of the other aspects of this that gets difficult is that the firmware versions that run underneath the networking world and with the SANs and servers—that all changed on its own metronome and sometimes incompatibilities occur.”
The advantage to operating with hyper-converged infrastructure according to Foley is that it is easy and it can be treated like an appliance.
Appliance approach
“The disadvantage is really to do with the issue of ‘management scale.’ In the past in enterprise systems, we typically tried to do what’s called ‘out of band management.’ What this means is that we would access a system out of the main production band,” he said.
“Basically, what we would try to do is isolate the management infrastructure from the production servers so that a management network failure or management application failure doesn’t bring down your production.”
Hyperconverged infrastructure is ‘in-band management’ so the management of the environment is in the box. If something happens to the box and it goes down, management goes down as well.
“And that’s OK for small installations, it’s a price worth paying for the convenience. But for large installations, it’s a different story.”
Ltd
