Posted by: Jordi Garcia, Associate Professor, Universitat Politecnica de la Catalunya, CRAAX Lab.
Fog computing has emerged as a novel and challenging technology, which aims at complementing the features of the cloud computing paradigm by exploiting the advantages of locality. Moving some computation to the edge of the network is especially attractive in a smart city scenario, where a vast network of heterogeneous resources is widely distributed all over the city. In such scenario, using the available fog resources allows the deployment of a rich family of new services that combine the advantages of both technologies, cloud and fog. In fact, the amount of computing load that can be moved down to the edge will actually depend on the edge devices capabilities. Therefore, some key questions arise, for example, how much computing capability is there at the edge? Is there enough capacity to execute efficiently any deep computing application? And the most relevant one, could the cloud layer be ignored in case of sufficient fog resources?
Figure 1 – Barcelona overview
For this reason, we have estimated the potential fog computing capabilities in a real smart city: the city of Barcelona. To do this, we have analyzed the available statistical information about Barcelona and estimated the number of computing devices in the city that could be part of an eventual fog environment, including personal computers, tablets, smartphones, on-board car computers, WiFi routers and some other embedded computers deployed as part of city services, such as connected traffic lights, traffic panels, street cabinets, and some other devices. From these estimations, we have calculated that there is a computing potential of almost 20 million of billions of instructions per second (19.2 peta IPS). It is worth emphasizing that this estimation is just a lower bound due to the limitations and inaccuracy in the available information sources. In addition, the technology deployment on society is constantly expanding and these figures will be rapidly and unavoidably increasing.
Figure 2 – Illustration of our envisioned smart city scenario: In such a scenario, the application could be offloaded for execution ① on a virtualized cluster ②
In order to understand the magnitude of these figures, let’s assume a computationally intensive task in the field of computational sciences (quantum mechanics, weather forecasting, climate research, molecular modeling or physical simulations) requiring, for instance, one year of execution in a standard personal computer. Such a task could be executed in the fog cluster of Barcelona, in theory, in less than 20 seconds. Or alternatively, we can compare this capacity with the cloud infrastructures of a major cloud services provider, for instance, the Amazon Web Services (AWS). Notice that Amazon, as other competitors, has historically been secretive about its data center locations and capacities; however, browsing through Internet one can easily figure out the orders of magnitude of their data centers. Amazon has a network of data centers spread around the world in order to maximize coverage and minimize failures effects, and we have found that Amazon proposes having less than 100,000 servers per data center. According to this information, the fog cluster of Barcelona is from 10 to 20 times as powerful as the largest Amazon cloud data center. But Amazon has a wide deployment of data centers around the world. According to the same source, they suggest the total number of servers to be at least 1.5 million, but could range as high as 5.6 million. With these approximate figures, an eventual fog cluster of Barcelona would have the same order of magnitude of computing capacity than the accumulated set of all Amazon data centers. However, note that this last comparison is not proportional because the AWS data centers are covering all the World and our estimation focuses only to an area covering just the city of Barcelona.
Based on the presented analysis, many relevant questions come up, for example, what is the potential fog computing capacity in the whole world? It seems evident that this measure should be definitely much larger than the accumulated capacities of all major cloud service providers. Hence, from a conceptual perspective, one may wonder whether cloud service providers will be essential in a near or mid-term future. The answer to this question may become the seed for novel business opportunities.