1
Introduction to Edge and Cloud Networking

1.1. Introduction to digital infrastructure

For the next 10 years, digital infrastructure in Cloud Networking will establish itself as the basic standard. This standard has been adopted by all network and telecommunications equipment manufacturers. It consists of four elements: the terminal equipment, an antenna, an optical fiber and a data center. To understand the reasons that led to this architecture, we must start with the basic element: virtualization.

The virtualization process is described in Figure 1.1. This process is a result of moving from a physical machine to a logical machine. The first step is to write code that does exactly the same thing as the physical machine. Assuming the physical machine is a router, the virtual router code must perform the same routing and send the incoming packet processed by the logical code on the same outgoing line as the physical machine would.

The next step is to compare the performance of the physical machine and the logical machine by running it on the processor of the physical machine. Without accelerator hardware such as ASICs (application-specific integrated circuits) or Field-Programmable Gate Array (FPGAs), performance will easily drop by a factor of at least 10 and possibly as much as 100. If we assume this loss by a factor of 20, it would take a processor 20 times more powerful to achieve the same performance, which is not a problem with data center power. However, since energy consumption is very roughly proportional to processor power, it jumps to a high level.

The next step is to try to minimize the energy expenditure. To do this, the processor of the physical machine supporting the logical machine must be occupied as close to 100% as possible. As this is not really possible, we must try to stay around 80%. To achieve this, a sufficient number of virtual machines must be multiplexed to achieve a very good CPU utilization.

Figure 1.1. The virtualization process.

The solution is to group virtual machines so that there are exactly the right number of them. If demand is too high, virtual machines must be migrated to other servers and vice versa to maintain high CPU utilization. We can also see from Figure 1.1 that data center utilization is the solution since the many servers are either put into sleep mode if not in use or they run at high utilization. Optimization of energy consumption is therefore achieved by migrating virtual machines so that all servers not in standby mode are heavily used. Virtual machine migrations, that is, the movement of virtual machines from one server to another, are in the vast majority of cases carried out in the same data center and much more rarely between separate data centers.

Figure 1.2 shows a data center with its virtual machines. As shown, there are continuous migrations to optimize operation. We also need to be able to give the virtual machines the power they need to perform the requested task. To do this, we need an orchestrator of the data center resources that are allocated to the virtual machines.

This software virtualization should be replaced gradually by hardware virtualization because of reconfigurable processors, but it will take many years before this new generation arrives, which will consume much less energy and greatly increase performance.

Figure 1.2. A data center and its virtual machines.

The question arises as to which physical elements can be virtualized and which cannot be virtualized. In fact, it is better to look at the second part of the question since everything is virtualizable except for three elements: the sensors, the wireless communication cards and wired communication cards. Sensors are not virtualizable because they have to capture something, which cannot be done by a code. For example, we cannot measure the temperature in a room by writing a code. In the same way, we cannot capture an electromagnetic signal with a code, nor can we always send a light in an optical fiber by a code. Otherwise, everything is virtualizable: a Wi-Fi box, a firewall, a key, a switch, etc.

Cloud Networking is precisely the network solution that uses the digital infrastructure that was described at the beginning of this chapter, that is, based on four elements: the terminal equipment, the antenna, the optical fiber and the data center. We will start by describing a few types of Clouds and their importance.

The Cloud is above all a mechanism that consists of grouping the resources of a company in the Internet rather than having them directly in the company, in order to share them with other users and benefit from a strong multiplexing of the resources and therefore a reduced cost. Cloud providers also benefit from multiplexing by selling shared resources to users who may be located on different continents.

In the early 2000s, the utilization of hardware, software and personnel resources was not optimized, since these resources were heavily used only during peak hours and hardly at all at night. Average utilization calculations showed that resources were used at less than 20%. By connecting several companies to the same common resources at different peak times, it is possible to achieve utilization rates of around 80% without increasing the resources.

Figure 1.3. Virtualizable and non-virtualizable devices.

The problem that immediately arose concerned the data of companies that are in a public Cloud and are therefore often at the mercy of attackers or states requesting information from their providers for cybersecurity reasons.

Private Clouds have been democratized to take this issue into account and have become the majority. The data is often installed on several private clouds within the framework of either large companies with several sites or companies with a single site but independent departments.

Today, there are different types of Clouds that have become more complex to accommodate new diversification and availability parameters needed by businesses.

The first type concerns distributed Clouds. As shown in Figure 1.4, these are different types of Clouds offered by the same provider: public, private, close to the user, on the Edge, or in the core of the Internet network but much further from the user, which we will call the Core Cloud.

The Edge (data centers on the edge of the core network) or Cloud (data centers inside the core network) provider can offer several types of services that we will detail later: an infrastructure, a platform or an application software.

Figure 1.4. A distributed Cloud.

Another widely used term is Hybrid Cloud, in which the data centers can be both private and public but can come from different Cloud providers. The hybrid Cloud is therefore a solution that combines a private Cloud, one or more public Cloud services and often proprietary software that enables communication between each service. By opting for a hybrid Cloud strategy, companies gain greater flexibility by shifting loads between different Cloud providers as needs change, and costs can vary rapidly.

An illustration of this type of Cloud is provided in Figure 1.5, where we see the connection of the two Clouds realized by access to multiple applications carried by the public or private part.

Other types have also been defined, such as multi-Clouds, which bring together several providers to support all the services requested by companies and which allow better availability in the event of overload of one of the Clouds in the environment. These multi-Clouds bring together both public and private Clouds and different types of services, platforms, infrastructure and applications.

Finally, the term omni-Cloud is the most general to take into account the multitude of possibilities of associations and structures of Clouds.

Figure 1.5. A hybrid Cloud.

Figure 1.6. Hypervision and containerization.

In Figure 1.6, we describe the internal architecture of servers inside a data center. There are two main possibilities: hypervisor and containerization. The first is the older one, it concerns the support of virtual machines as it was originally conceived. The second solution is gradually replacing the first with a simpler, less expensive and more flexible architecture.

Hypervision consists of using a hypervisor on a standard physical machine (commodity) which is a software able to support several virtual machines simultaneously through one or several operating systems (OSs). The hypervisor supports domains formed by an operating system and the virtual machine running on it. The Domain 0 or Dom0 is specialized in processing the I/O of the other domains on the base physical machine.

There are different types of hypervisors. Paravirtualization requires that the operating systems be slightly modified so that all the processing requested by the virtual machine can be done natively on the basic physical machine. On the contrary, the second solution is to accept the operating systems without modification but with the introduction, above the hypervisor, of an emulation software able to adapt the execution of certain functions to the underlying physical machine.

Containerization is gradually replacing hypervisor with a division of services into microservices that each run in a container. In this case, a single operating system is used that supports...