Читать книгу Understanding Infrastructure Edge Computing - Alex Marcham - Страница 58

As mentioned previously, the internet is far more than the combination of physical infrastructure and logical protocols; it is a network of networks that are joined as much by agreements between their operators towards a mutual benefit as they are by anything else. This can be best seen in the way in which the networks of different operators interconnect and exchange data. This exchange of data between networks is essential; without it, endpoints of one network would not be capable of accessing resources or endpoints on another network, making the internet we know today useless.

The large‐scale exchange of data between networks typically occurs inside an internet exchange (IX). Often, IXs are located within large data centres which terminate a high number of WANs inside the facility. The higher the number of networks aggregated into one location, the higher the theoretical value of that IX as a place for networks to exchange data. Chapter 7 focuses on this interconnection.

The physical locations of IXs are of considerable interest when thinking about infrastructure edge computing. Our previous example describes only the location of the IX itself, and not the locations of any of the endpoints which are ultimately sending and receiving the traffic. Consider the scenario, however, where these two endpoints are located close together but neither is near to an IX. In this example, each endpoint is connected to a different access network, and so for the data that they are sending each other to reach its destination, it must go all the way to the IX, be exchanged between their networks in that remote location, and then it will be sent all the way back to those endpoints.

Where the endpoints are using applications that rely on either very low latencies between endpoints or are sending a large amount of data, this is not an ideal network topology. For the former, latency is added between the source and destination of the traffic due to its need to traverse the IX before it can be delivered to its target endpoint, which would be minimised if a point of data exchange were available closer to those endpoints. For the latter, network resources are used between each of the endpoints and the IX to transfer the traffic, which some entity (even if it is not the users of each of the endpoints themselves directly) must pay for, with the possibility to introduce congestion as well.

These factors have led to the need for more distributed points of data exchange across the internet. Infrastructure edge computing addresses this challenge by the implementation of an edge exchange (EX) within infrastructure edge data centres. Conceptually, this idea is simple; a smaller‐scale version of the functions performed by an IX can be implemented inside an infrastructure edge data centre to allow networks to exchange data at locations closer to the endpoints, which are generating the traffic in order to decrease latency and minimise the cost of data transportation. This topic will be explored in a later chapter in this book, but its existence is a useful primer for the reader at this point as well.

Many agreements to exchange data between networks are reciprocal. One network may agree to interconnect with another because each network would bring a roughly equivalent benefit to the other, in terms of accessible endpoints and resources as well as establishing a balance in terms of the volume of traffic that each sends and receives with the other network, creating an equilibrium.

However, this understanding has been tested in recent years by the widespread use of streaming video services and other networks which send orders of magnitude more traffic than they receive. These services can place significant strain on the networks which they interconnect with due to the sheer bandwidth use generated by the widespread adoption of these services. This has prompted some network operators to move such relationships away from traditional reciprocal agreements.

Other agreements are considered “pay to play” arrangements, where a network may agree to interconnect with another network only if that other network agrees to pay for them to do so. Future payments may then be organised on a usage basis or on some other means as agreed between the parties involved. Although unpopular with some, the idea of paying for interconnection is a viable option where one party requires an additional incentive to interconnect with another. Alternatively the interconnection may not occur, reducing the overall ability of the internet to grow over time.

On‐ramps are an interesting part of network interconnection and exchange which have become more prominent over the past decade with the rise of cloud computing. Although there is some variation in how the term is used, an on‐ramp generally refers to a dedicated piece of network infrastructure which gives one or more parties direct access to the network of a cloud provider. Examples of these on‐ramp services include Amazon Web Services (AWS) Direct Connect [3] and Microsoft Azure ExpressRoute [4]. As more applications operate from cloud instances, such services become increasingly important to minimise latency and the cost of data transportation long term.