The Challenge of Bringing Intelligence to the Edge

Feb 15 2019

NGD Systems

Clearly, edge computing use cases are important use cases, and creating and deploying edge computing solutions for these types of problems has significant complexities. The number of important use cases for edge computing are also increasing rapidly as instrumentation (sensors and video cameras for example) rapidly increase. However, there are significant difficulties in fielding many of these solutions. Self-driving cars and aircraft engine management systems are great examples of the difficulties that many edge computing solutions must overcome. Both use cases have limited space and power available for the edge computing solution, as well as having significant thermal and cooling issues. While some edge computing problems do not have to deal with these limitations (industrial IoT applications being one example), most edge computing use cases cannot utilize standard datacenter equipment as part of their solution.

To put this problem in perspective, consider the amount of power utilized by a standard dual-processor 2U server with 1TB of memory, which is roughly 1000 watts (depending on the particular server model and configuration). This is about 35% of the maximum power that a typical gasoline-powered car’s alternator can produce. For an electric car like the Tesla3 with the long-range battery (75kWh), a 1000-watt server would consume about 7% of the car’s available charge during a 5-hour drive (roughly the car’s maximum drivetime at a speed of 55 MPH). Other edge computing environments have temperature and/or space limitations. For instance, typical datacenter servers expect inlet air temperatures between 50 and 95 degrees (Farenheit), require enough airflow to dissipate 4100 BTUs of energy per hour, and take roughly 1.15 cubic feet of physical space. Datacenter servers are also typically not engineered to sustain significant amounts of vibration. In an aircraft installation, these requirements can be very problematic to meet.

Then there is the application needs itself. While a 2-socket server may seem fairly powerful, it is probably not adequate for a large number of edge computing applications, and scaling a solution to multiple servers obviously becomes very problematic. This is how prototype self-driving cars have seen the power for their computers and sensors reach power requirements of up to 2500 watts. Alterative computing technologies such as general-purpose graphics processing units (GPGPUs) can certainly pack more computing power into a smaller volume than is possible with a standard X86 server, but they also tend to have their own significant power and cooling requirements (though less than datacenter servers do). In our next article, we will explore how a new technology known as computational storage provides an alternative path to address these issues.

Self-driving car that uses edge computing technology to navigate the road.

355 Goddard, Suite 200
Irvine, CA 92618

For Sales Inquiries:

For General Inquiries: