We are mid-way through 2021 and 5G networks are beginning to roll out across the globe. The technological goals are clearly stated and are on the roadmap to be implemented: Optimized use of old and new spectrum, enhanced mobile broadband, massive machine-type communication, ultra-reliable low latency; to name a few of them. The scale of investment in research, development, and new infrastructure is massive in comparison to previous generations, so the industry is making its biggest bet ever.
As service providers ante up, they appreciate the economic responsibility to get it right – accountable to the investing stakeholders, to national governments invested in a “win”, to internal teams looking for the lifeline to help them thrive in the digital era, and ultimately to the customers who pay for services and applications. This pressure is higher than in previous generations as mobile networks are no longer simply a tool to make phone calls. They will very soon become an essential fabric that will underpin every aspect of our lives.
Take the mobile network away and society will find it very hard to function at all. Stakeholder, customer, and user interests are as important as ever, so the final success of 5G will be determined by our ability to change the world while the world is constantly changing. And society will be the judge.
Needles and Haystacks
Now, 5G enters its next phase: the development of an ecosystem of services and applications – the things users want and pay for. But do not look for the killer application, as you may hardly find one. Instead, look at how society evolves and how fast technology adapts. If you look close enough, you will find societal evolution facilitated by advances in technology. The process here is iterative, and while at times the evolutional shift is being driven by society, there are other times it is driven by technology. Both sides then evolve together, continuously. It follows that the 5G ecosystem will grow inside this evolution.
Multi-dimensional variance makes it very hard to understand the pace and specific directions of this evolution. We can, however, look at major trends, analyze them, formulate a vision, and set a strategy to realize this vision. This seems like finding a needle in a haystack. In fact, it begins with finding the significant haystacks, or in our case the major trends. This is what we discovered:
- On the technology side, the disaggregation of the radio network along with the new service-based architecture for the core network changes the essence of mobile networks. The network elements become programmable and containerized. At the same time, the compute locations are extended from central data centers, through the edges, all the way to the user premise.
- On the society side, new ways of working gain momentum, partially due to the impact of Covid. Where possible, “work” is disconnected from an exact location (an “office”) and is distributed. This comes with a renewed interest in security and privacy.
- On the operations side, it is rather clear now that our current solutions for intelligent automation are far from sufficient to cope with the networks today. And this is before Machine Type Communications with latency and jitter constraints see wide-spread adoption.
Previous generations of wireless have been served well by the Internet of today, albeit with security and mobility management functions bolted on a system originally designed for best-effort content consumption. Multi-dimensional variance induced by 5G, as discussed above, seriously challenges the static constructs currently in place for the Internet. As we scale out our 5G deployments, and as 5G enabled experiences evolve from human centric consumption to machine centric consumption and collaboration, we see a need for simultaneous evolution of the Internet in order to harmonize the overall 5G ecosystem. Our vision for this evolved Internet is that of a globally distributed Grid which weaves together a fabric of heterogenous compute, data, and intelligence to enable context-aware, immersive, and distributed modern applications anytime, anywhere, on demand.
The Grid is driven by applications (hence the value of the 5G ecosystem) that require fungible resources to be allocated at the right place and the right time. The fundament of the Grid is a virtual data center delivering logical pools of adaptive compute, securely and timely, while spanning multiple geographies. Autonomic controls accelerate automation effectively turning the Grid into a self-organized and self-optimized construct. Producers and consumers can then not only book and utilize capacity and resources, but also extend the Grid by adding new or handing back spare capacity and resources.
Building a global Grid is no small task. It takes a village. This is the reason we have co-founded the Open Grid Alliance (OGA) in partnership with other industry trailblazers to realize this vision.
Circles and Squares
Low latency, high performance, distributed applications seem lofty goals, when faced with imminent challenges for secure and scalable deployments of disaggregated containerized network functions distributed across a broad spatial scale, as promised for 5G. We clearly need to square a few circles, namely (1) programmability vs. time-sensitivity, (2) real-time QoS vs. reliability, and (3) local autonomy vs. global objectives. And do not forget the applications!
Applications are time-sensitive on a range of time aspects and time scales. Some require establishing a connection in a certain time (e.g., mission critical services), some need guaranteed latency in milliseconds, some depend on bounded packet delay variations, and some must process and transport different types of data in milliseconds, seconds, or minutes. With disaggregated, virtual, and programmable network elements, from RAN to Core, changes can and will happen regularly and frequently, with short notice or no notice at all.
- We need to pair the Grid with a time-sensitive fabric, to cope with this dynamicity, calculate requirements on-demand, and acquire and release resources upon an application request. At any time. Between any given end points.
With rapidly changing network conditions, managing delivery of highly demanding next generation applications means the network will have to respond dynamically, from the time of request up to the end of the session. And because no resource is unlimited, we will be constraint by cost and shared resources. We should not limit cost considerations to CAPEX and OPEX alone. Instead, we want to look at the environmental footprint and impact as well.
- Next, the Grid will evolve with a layer of intelligence and optimization to maintain real-time QoS, with built-in reliability driven by application demands.
Many stakeholders play an active part in network and service operations. Classic models of centralized control have shown to not scale. Naïve approaches to delegate control are often not able to evolve fast enough when conditions are shifting. The other extreme, unchecked and fine-grained autonomy, can easily result in operational chaos. Coordinated and distributed control loops can help us to automate regulative, monitoring, and repair activities in a dynamic environment. Cybernetics and cognitive science have plenty to offer: including Wohl’s SHOR, Rasmussen’s SRK, Klein’s RPDM, and Boyd’s OODA.
- To reconcile local autonomy and decision-making with global objectives, the Grid will require an autonomic layer. This layer will provide autonomic capabilities along with distribution strategies for automatic, autonomous, and autonomic operations.
Once the Grid is established along with the layers for intelligence and autonomic capabilities, we can look at distributed modern applications. There will be a lot to discover here, insights and invariants. Some we might know already, just in another context. Some we will re-discover. And some will be new. We do know already that distributed modern applications will have a time budget, labeled as latency or jitter or speed of up-and-downlink, that spans over a dynamically variable scale. The Grid will enable such applications by stitching together the constituting software workloads that may be stretched across a scalable edge-to-cloud continuum, while maintaining strict latency and jitter constraints dictated by the application.
Multi-Cloud Services Grid
We are taking the first step towards implementation of this continuum by building a Multi-Cloud Services Grid (MCSG). MCSG is a multi-cloud framework that enables intent driven service creation and consumption with continuous optimization of underlying resources. And it does so by leveraging Kubernetes based control plane to stitch distributed modern applications with resources that may span across the central cloud, the near edge (e.g. metro data centers) and the far edge (e.g. micro data centers), and in turn may be owned and operated by their respective public and private cloud platforms. The MCSG is a joint collaboration of VMware and Vapor IO, and is baselined on a software substrate formed with integration of VMware Telco Cloud Platform with Vapor IO’s Kinetic Grid platform.
To use the metaphor from earlier in the blog, MCSG will square many circles which will enable three important advancements:
- Specialized Infrastructure – Fulfilling a service provider need, MCSG will unify specialized resources. In terms of hardware this can be ASICs or FPGA, including smart NICs. In terms of software this can be specialized versions of operating systems. Resource allocation might prioritize full reservation over common lazy allocation, especially for strict latency requirements. MCSG will be able to provision infrastructure in clusters with appropriate specialized support if required.
- Multi-tenant Services on Single-tenant Platforms – Kubernetes, as examples, comes with identity and access management as well as isolation out of the box but targets single-tenant enterprise situations and SaaS is designed for multi-tenancy but targets single service situations. MCSG will reconcile the two worlds.
- Distributed Control Plane – MCSG will separate the application from the underlying infrastructure and respective management & control software. This separation allows us to distribute the control plane as well as the application to a place where they are best situated: in a specialized cluster, any public cloud, any private cloud, and any location -from central data center to far edge or at the user premise.
These three advances facilitate what we call ‘hypercomposed’ applications – dynamically composed applications with the ‘right type’ of resources in the ‘right amount’, in the ‘right place’, and at the ‘right time’, with intelligent optimization across all four dimensions at all times.
This is an exciting journey. And MCSG is only our first concrete step toward broader OGA ambition for evolving the internet with an Intelligence Grid. We are contributing the MCSG to the OGA – a member-supported collaboration organization that produces vendor-neutral strategies for Grid topologies – and we invite you to do the same, by contributing your relevant assets for co-creation at the OGA. We are looking forward meeting you at the OGA!
Read more about the Open Grid in this blog.