In my last blog post, Edge Native Applications Are Conquering the Smart Device Edge, you were introduced to edge native applications and to the categorization of the various edge tiers based on the LF Edge definition. This blog post will focus on the hardware and software aspects of each categorization to give you an extensive understanding on all possibilities at the edge – from every angle!
InfoQ roundtable with world-class edge practitioners – July 19th
InfoQ is hosting a special round table about “How to Successfully Build and Deploy Applications on the Cloud.” If you want to join this interesting discussion, register here.
Edge != Edge
It’s pretty clear that when people are talking about “the Edge” they can be meaning different things. Now, let’s take a closer look at each edge category, and focus especially on the hardware and software parts. In the overview graphic below you can see where devices from each tier are located. Devices in the constrained device edge are usually very small with nearly any form-factor, and are highly distributed anywhere in the field. You can find a proper definition on constrained devices in the RFC7228.
In the smart device edge category you can find more powerful hardware, but which is still constrained when it comes to CPU, memory, and storage. These devices are usually located outside of an office building, whereas in the on-prem data center edge, you can find standard rack or blade servers for general computing, storage, or operating a virtual desktop infrastructure (VDI). The equipment in that category is usually owned and operated by the enterprise before the last mile. Then, in the service provider edge tier, the service provider is operating the infrastructure. An example could be multi-access edge computing (MEC) which is specified by ETSI.
Software & Hardware Overview at the Edge
In the table below you can see the differences of each edge category from various angles. You can see that the constrained devices do have a very small form factor, and are used for very specific tasks. In this area, tinyML is on the rise as it brings machine learning capabilities to ultra-low powered hardware, even to sensors. Devices in the smart device edge are getting more powerful and less expensive, leveraging even GPUs in some use-cases. The use of so-called edge native applications is increasing especially on these device types.
It also depends on the used hardware which programming language to use. For example, running a Java application on a low-powered embedded hardware can be more challenging, and would take longer to start, than the same application written in C++.
From a developer point of view, devices from the on-prem data-center edge to the regional edge are very similar. The use of virtual machines and containers are standardized, and the CPU power, memory, and storage is sufficient for most tasks. Just if there is a need for data-intensive computing (e.g. training machine learning model) there can be limitations which the developer needs to be aware of.
All-in-One table: Comparing the general properties of the different edge tiers. On the top you can see example devices for each category.
I hope this overview gives you a good understanding about the different edge tiers!
- IOx Edge Compute Overview
- Edge Compute Learning Track
- Edge Native Applications Are Conquering the Smart Device Edge (Blog Post)
- Edge Native (blog post)
- Cisco Edge Intelligence
- Cisco Hyperflex Edge
We’d love to hear what you think. Ask a question or leave a comment below.
And stay connected with Cisco DevNet on social!