Dealing with the growing problem of edge and IoT sprawl
Amidst all the hype regarding the massive growth potential for IoT and for edge computing in general, there is an elephant in the room, and that’s edge sprawl.
It’s as if the edge evangelists are happy to boost the use of IoT and edge computing, but don’t want to be bothered by any adverse consequences of their actions. The problem is that for early adopters, in areas such as retail, those consequences are already showing. For example, I was talking recently with Andy Brewerton, an industry acquaintance of many years who’s now COO with edge HCI developer Sunlight.io, and he mentioned that there’s fast food restaurants in the US with more than 20 different local systems installed.
Do those systems – from point-of-sale through CCTV to aircon controls – all count as edge? Some might argue not, yet they are all outside the data centre, they exchange data with HQ, and it’s very unlikely that there will be IT support capability within each store.
The in-store/in-office IT closet is getting crowded
At the moment, most – if not all – of them will be discrete systems with their own local hardware, whether that’s an embedded controller or an appliance based on something like a Raspberry Pi or Intel NUC. That’s a lot of hardware to maintain and it’s a lot to find space for, to the extent that there’s anecdotal reports of system installers arriving on-site and finding it hard to make room for their latest piece of kit – and then of course there’s the challenge of finding spare power and network feeds for it.
It’s a problem that is only getting worse. Think about the desire to instrument as many devices and processes as possible, so they can report back in real-time as part of the IoT, and the growing use of AI-based computer vision and machine learning. Those are huge drivers behind the growth of edge computing, a growth that shows no sign of slackening off any time soon.
When server sprawl hit critical mass in the office and data centre, we mainly responded in one of two ways: we either relocated and centralised the offending systems, or much more often, we virtualised and consolidated them.
Could that work at the edge? Centralisation can’t, because there’s reasons these things are at the edge – it’s where they need to be! But there is growing interest and activity around edge consolidation, using much the same techniques we use elsewhere, such as virtualisation, containerisation, ‘edge cloud’ and hyper-converged infrastructure (HCI).
The need for more power, but a tighter focus and lighter weight
Of course, the actual technologies used need to differ: in this constrained environment there’s no room – or need – for the kind of full-function hypervisors or container platforms we use in the data centre, for a start. So this is where lighter alternatives play, such as Sunlight.io’s Nex hypervisor (so-called because it derives from Xen and is Xen-compatible), Zededa Eve-OS, Scale Computing’s HC3, Edge HCI from SourceCode, and the K3s lightweight Kubernetes distribution, to name just a few.
Conversely, edge consolidation needs a somewhat bigger hardware platform – not a full-function server, of course, but certainly more than a NUC or RasPi. As Andy pointed out in our discussion, it’s unlikely that you’ll be able to consolidate multiple workloads on a single RasPi, but edge HCI will let you consolidate four RasPi workloads onto one larger device – he highlighted solutions such as Lenovo’s compact and rugged ThinkEdge SE30.
And while a bigger, more expensive device may not look like a win at first, it’s not just the hardware deployment and management that it simplifies. Andy noted that virtualisation and HCI also simplifies software deployment for Sunlight.io’s ISV partners, plus the more-capable hardware platform gives a degree of scalability and future-proofing.
Given that edge sprawl is already happening, and is not sustainable, something needs to change. What’s your preferred route forwards – or have you not thought about it yet?