Enterprises are actively moving applications to “the cloud” – which usually means public clouds in hyperscale data centres. And the drivers for this include minimising capital expenditure and maximising flexibility and scalability. But using a centralised cloud does not address all application demands, which include low latency, standalone resilience and data sovereignty. A successful approach must include centralised clouds (whether public or private) as well as onsite micro clouds.
Drivers for centralised clouds
Today, the most popular model for cloud deployments is centralised public clouds. And there are some very good reasons for this:
- Low capital expenditure: Public cloud pricing is based on a consumption model. You pay for what you need while you use it. And there are no upfront costs for buildout.
- Scalability: The enormous size of public clouds means that you can scale up your deployment as needed to handle seasonal or time-limited increases in demand.
- Flexibility: Public clouds typically have deployments around the world, enabling users to host applications near where they are needed and move them as required.
- SaaS models: The reasons listed above make public clouds ideal for hosting software-as-a-service (SaaS) applications, which need the flexibility to scale up and down as per demand.
Drivers for edge clouds
However, the benefits listed above may be outweighed by other considerations that make centrally hosted clouds unsuitable. Here are some of the drivers for moving cloud resources to the customer edge:
- Low latency: Some applications require low latency access to compute resources that cannot be met by centralised clouds. Examples include private 4G and 5G, internet of things, augmented reality, virtual reality and smart manufacturing.
- Reduced backhaul: Applications such as video surveillance produce tremendous volumes of data, only some of which contain useful information. By processing the video stream locally, the required uplink can be dramatically reduced because only alerts or aggregated data are sent upstream.
- Data sovereignty: Some companies, industries and jurisdictions have requirements for keeping data local. A centralised cloud requires transporting data out of the specified area and so it is not an option.
- Standalone resiliency: What happens to your mission-critical applications when the network links fail? If the applications are hosted centrally, you are dead in the water. Local hosting provides the ability to keep going until the link is restored.
- Lower costs: A centralised cloud can be very expensive for applications that run 24×7. For example, the cost on Amazon Web Services (AWS) of reserving a single “a1.xlarge” instance (4 vCPU, 8GB memory) is $46.94 per month, or $1690, over a period of three years. To enable a similar environment on a universal customer premises equipment (uCPE), four CPU cores and 8GB of memory are required to be added. That means a one-time incremental cost of about $800 – less than half the three-year cost in AWS.
Edge cloud is different
The growth in demand for edge computing means that clouds must be built in customer edge locations. However, these edge clouds are intrinsically different from centralised hyperscale data centre clouds.
- Scale is different: Edge cloud means that telcos or enterprises are managing thousands and tens of thousands of clouds at different sites. This requires centralised management and orchestration that can handle the scale.
- The environments are less uniform: Hyperscale data centre clouds are uniformly planned and controlled. It is not so for edge clouds, which are being built in aggregation sites or customer edge locations. And these clouds must be deployable over a wide range of power/heat and operating environments.
- Edge clouds require resilient and secure management access: Physically accessing edge clouds isn’t always easy. It may require scheduling or customer approval. Edge clouds must support remote management that is secure, is able to operate over multiple fibre and wireless networks, and is resilient in case of hardware failures.
- Installation is not done by one team: Entire clouds/cloud networks must be launched remotely using zero-touch provisioning (ZTP) so that installation can be handled by non-technical people. Cloud cabling should be simplified to the point that it is difficult to create cabling mistakes.
Broaden your thinking on the cloud
Whether you are an end user or a service provider, the cloud is an essential part of your strategy. But you must consider all relevant requirements and options. And as you include edge cloud in your plans, be sure to think about how you can successfully address the special needs of these deployments. Partners like ADVA can support you with products and planning to help you achieve edge cloud success.