Although Edge Computing has emerged as a “new” technology trend, in reality, we’ve been deploying compute outside of the data centre for decades. Banks and retail organisations have used intelligent terminals and distributed architectures for local processing and subsequent data centralisation. As Edge has become more attractive to businesses, we need ways of implementing efficient IT infrastructure at a low cost in a range of diverse scenarios and locations. Scale Computing recently announced the HE150, an edge-focused cluster based on Intel NUC hardware. We look at how this could be the model for edge computing, not just because of the hardware, but through the HCI software the solution runs.
It’s no secret that I’ve been a fan of the Scale Computing HC3 platform since the first presentations at Storage Field Day 5 in April 2014. What piqued my interest were the lightweight and elegant features such as SCRIBE and the integrated state engine.
- Scale Computing Debuts HC3 in Google Cloud Platform
- Scale Computing Moves Deeper into the Enterprise with All-flash HCI Nodes
Rather than run storage as a set of virtual machines across physical nodes, SCRIBE and the other features of HC3 are directly integrated into the operating system. This results in much lower compute/memory overhead than other solutions. While HC3 is arguably not as fully-featured as say, VMware vSphere, over the years, support has been added for cloud-based replication and all-flash nodes.
Perhaps one of Scale’s greatest strengths is support. I currently run two clusters; an older HC3 with HDDs, which is in my co-lo lab and a new HE150 cluster based on three NUCs (more on that in a moment). In both scenarios, the deployment and installation process was seamless. Similarly, for the one drive failure I’ve experienced, Scale quickly shipped a new drive that was replaced by the “intelligent hands” team at the co-lo.
It might seem that I’m gushing a little about the support process, but as we will discuss, this asset is an excellent strength in supporting edge environments.
So, what is the HE150? Scale currently has two edge computing solutions, the HE500 and HE150. The 500 series is rack-mounted, uses Intel Xeon E-2100 processors with between 16-64GB of RAM. The five family models offer a mix of HDD, hybrid (HDD & SSD) or all-flash configurations and a combination of either 1Gb or 10Gb Ethernet. Prices start at around $5,500 per node.
The HE150 is an even smaller configuration, based on the Intel NUC. The NUC is a small form-factor PC running an Intel Core processor, M.2 SSD storage, up to 16GB of DRAM and Gigabit Ethernet. An entire 3-node cluster is a cube of sides less than 12cm each (see figure 1).
In my implementation (as shown), I use a standard Gigabit unmanaged switch to interconnect the nodes. All HC3 clusters have either a physical or virtual network backplane. In the case of the HE150, this is implemented as a shared network on the same infrastructure as the user connections or via direct links between each node. In this configuration, user connection is over Wi-Fi. The Tech Field Day 20 presentations from Phil White go into much more detail on the networking configuration and are well worth your time.
The requirements for edge computing are diverse and wide-ranging. This can include computing infrastructure in retail branches or inhospitable locations such as oil rigs, wind farms or water treatment plants.
In pretty much all of these scenarios, efficiency is a critical success metric. With (for example) 10,000 edge nodes to deploy in branch operations, excess capital costs soon multiply. Each location needs just the right amount of computing. The NUC form-factor occupies very little space while making the “unit of support” a single node. Failing nodes are simply replaced.
Equally, engineer site visits need to be as infrequent as possible. The HE150 form-factor makes the support process simple and could be entrusted to any local employee. When field replacement is at a node level, new hardware can be shipped and replaced by onsite staff. Site visits occur for only the more complicated or challenging problems.
As I mentioned earlier, the Scale Computing support is excellent and should scale (no pun intended) to meet the requirements of customers. With literally tens of thousands of nodes in the field, the support model will be critical to the success of Scale’s business.
As a hardware solution, HE150 NUCs are both cost and space-efficient. But what about software? We’ve already seen that on existing platforms, HC3 is lightweight and easily managed via GUI or API. The same HC3 software is used on HE150 and HE500 installations. The HCI overhead is small and doesn’t consume too much of the cluster resources. Each HE150 deployment is supported centrally and managed through APIs. You can learn more on this Storage Unpacked podcast we recorded last year with CTO Phil White.
Although HE150 is targeted at edge use cases, any non-data centre requirement could be met with a three-node cluster. This could include home or test labs without traditional infrastructure or providing developers with local test clusters. The benefit to the Scale HC3 design is that applications will run on any cluster configuration, big or small. It’s just a case of how many VMs can be supported. As a result, we shouldn’t see HE150 as just for edge cases.
The Architect’s View
Over the years I’ve found my HC3 cluster much easier to use than the 5-node VMware vSphere cluster I also maintain. It’s probably the simplicity of the GUI that’s helped. I can spin up a VM from a cloned image or PXE boot from a range of Linux distributions and have a new machine ready in less than a minute. While I can do this on the vSphere cluster, the Scale cluster is just more accessible.
Unfortunately, my older cluster doesn’t support the HC3 API. The good news is that my HE150 environment does offer API support. As a result, I’m working on a series of posts looking at the practicalities of API automation, which is also applicable to edge configurations. You can find the first post in this series here.
As an edge solution, HE150 looks extremely attractive and has an aggressive price-point at around $5000 for a 3-node configuration. Of course, other HCI vendors could replicate the hardware model, but the question remains as to whether their software implementations are lightweight enough to make NUC-based edge computing viable.
Disclaimer: Chris was invited to Tech Field Day 20 with GestaltIT covering travel and accommodation. There is no requirement to blog or produce any content from the event and no content is vendor-reviewed before publication.
Post #BBF7. Copyright (c) 2020 Brookend Ltd. No reproduction in whole or part without permission.