Application Infrastructure for Multi-Cloud and Edge Computing

Perspective on 5G and Edge Computing

Applications are increasingly going to process more data and make decisions closer to the edges for better real-time user experience, compliance, etc. Gartner estimates that by 2025, 75% of enterprise-generated data will be outside of a central data center or cloud.    Use cases across verticals need geo-distributed application architectures – autonomous vehicles, digital healthcare, smart retail, smart cities, and industrial automation are examples.  5G is an important enabler for edge computing. In addition, a confluence of factors are coming together that will enable Edge (geo-distributed) computing. 

1. Network speeds – 5G offers 10x-20x higher speeds (multi-Gbps) and lower latencies when compared to 4G with improved reliability. This enables a better experience for distributed applications. 
2. Compute costs – Specialized hardware such as GPUs, TPUs have now become more affordable to be used at the edge. So one could envision offloading a compute-intensive task such as ML closer to where data sources are. 
3. Distributed cloud and datacenter footprint – The large cloud providers themselves are expanding their footprint. In addition, companies such as  Equinix Metal, Vapor IO, Cox Edge are building micro data centers and services. 
4. Modern application architectures that lend well to distribution – Many enterprises have focused on adopting microservices-based application architecture, where components are loosely coupled but tightly connected. This framework will help in distributing applications. 

What are the challenges for edge computing?

The key motivation of edge computing is to offload compute from the cloud to edge to process data closer to the source and make real-time decisions. So based on the use case and application, the edge could just have an ephemeral lambda function or in many cases a long-lasting footprint to run ML workloads, analytics, and even a local database, storage for compliance and efficiency purposes.  The key challenges with edge computing include:

1. Heterogeneity and scale – Unlike a cloud environment, edge environments exhibit heterogeneity w.r.t infrastructure capabilities, footprint, and providers, and there can be multiple locations. So applications need to be developed so they can be deployed in such heterogeneous environments. 
2. Connectivity and security – One of the consequences of heterogeneity is connectivity and security challenges. Edge locations can be connected either with wired or wireless networks or both. Depending on the provider and capabilities, network services such as VPN, firewall, load balancer, etc. could differ at each location. With the application footprint now spread across locations, the security exposure and attack vector increases too. Service chaining could be potentially used, but that increases the complexity with multiple locations – especially when it needs to be done reliably in an automated way. 
3. Data – There can be multiple data sources such as IoT sensors or servers and applications themselves. Paradigms for data collection, representation, and abstraction, secure streaming, and exchange between different services are required for distributed decision making and to have a feedback loop across the application span so the application can be adaptive.
4. Observability and Resiliency – Fault isolation for such a geo-distributed application can be hard, especially if done manually. Rich real-time observability is needed from infrastructure to application across locations along with a feedback mechanism that can be automated to reduce error and build resiliency. 
5. Dynamism – The edge can sometimes be mobile or the data/users being serviced can be mobile. So optimization mechanisms need to be in place to dynamically move the workloads from one location to other based on cost,  latency, bandwidth, proximity, etc. 

The last decade has been about digital transformation and cloud adoption. If going to a single cloud was hard, distributing applications across multiple locations is harder – but inevitable to be on the cutting-edge. 

How to build the application environment?

It is important to build an application environment that is easily portable across locations, vendors, and providers and one that can scale well. See the picture in the 5G Magazine for a high-level approach. It has three key layers.

• Common connectivity and security fabric – This is the key layer that binds all the locations together. It has to abstract the underlying providers and vendors and provide a unified view for the geo-distributed application. It has to be able to make it simpler to discover and connect disparate entities such as VMs, docker containers, Kubernetes clusters, network and security services, etc.
• Common data fabric – This has to build on top of connectivity fabric and provide API-based access to application microservices for data collection,  secure streaming, and exchange among different services that may or may not be co-located.
• Application orchestration and management – This has to enable application definition and deployment across locations. It has to provide a  framework for observability, feedback, and policy-based mechanisms for resiliency, scalability, and lifecycle management.

Having such an application infrastructure framework will allow for the non-disruptive evolution of the application over time when new locations, providers, services, and components are added in due course.

Doesn’t open source offer the solution?

Open source has been a catalyst for innovation and democratized access. It offers multiple tools and technologies to build a distributed infrastructure. Open source also gives a sense of ownership and control (compared to dependence on a vendor’s solution) to the enterprises and their technical operations teams. It becomes both a  temptation and necessity to weigh the merits of different choices for a given component before picking one. An example would be choosing container networking or storage plugin. In both cases, multiple options exist and it requires technical depth to make assessments.  But as was evident in the edge computing challenges, a geo-distributed application infrastructure needs piecing together a complex puzzle that requires both breadth and depth of skillset. The complexity can be further appreciated in the context of security. Different components (containers, network, data message bus, application, etc.) have their own authentication and security framework. So managing a cohesive end-to-end security framework in such a distributed environment is quite complex.  That’s not to say open source does not have the answer, but not every enterprise has the wherewithal for a DIY operation. Most modern businesses are digital businesses and application delivery is central to their success. So making the right tradeoffs becomes important for successful business outcomes. A vendor-based solution that is open source/standards compliant may offer a practical approach for such enterprises. One of the reasons why things work well in a single cloud is because the cloud provider offers well-integrated core infrastructure (compute, storage, network,  security, observability). The same cannot be said if application straddles different cloud providers and this becomes harder as one goes from cloud to the edge. 

How about the cloud providers?

Cloud providers are building edge cloud services. AWS offers Wavelength, Google has Anthos, IBM has Edge Application Manager for example. Some have tie-ups with  Telcos to offer low latency network connections between locations. This could work well for some customers and their applications and might be their preferred option. The applications will be tethered to a particular cloud provider. However, edge offers more choices with micro datacenter providers and vendors building smaller and specialized compute environments. Some customers may want to leverage best-of-breed environments that best meet their application and business needs such as proximity to end-users, latency, capabilities, or cost. Such customers are better off exploring vendor-neutral solutions. 

Does fledge.io offer a solution?

fledge.io has been built to address this challenge. fledge.io offers a single unified cloud experience across different clouds, data centers, and edges that is intuitive, easy to use, and based on open technologies. Our solution provides the core application infrastructure that includes geo-distributed application orchestration,  consistent application connectivity with zero trust security, continuous observability, and telemetry-based data collection and streaming across cloud and edge. Essentially, fledge.io offers a public cloud-like experience for such geo-distributed applications and is a fully cloud provider / vendor-neutral solution. We partner with cloud and data center providers, and we will provide a seamless experience for customers as their applications straddle heterogeneous environments.