Like any type of large-scale computer systems deployment, the short answer to the question ” how should my fog computing implementation ” be “depends”. But as this is not a particularly useful answer, Cisco’s chief systems engineer and architect, Chuck Byers, gave an overview on Wednesday, at the 2018 Fog World Congress, about the many variables, both technical and organizational, involved in the design, care and feeding of a computer configuration in the fog .
Byers offered general advice on the architecture of fog computing systems , as well as a little deeper immersion in the specific areas that all cloud computing deployments should address, including different types of hardware, network protocols and security.
Options in the fog configuration
Computing in the fog configuration often includes several types of processors, making it a heterogeneous environment . The RISC / CISC CPUs , made by ARM / Intel, provide excellent performance of a single process and a high degree of programming capability. “They will always have an important place in the fog networks, and almost all nodes of this type will have at least a couple of cores of that kind of CPU,” Byers said.
However, they are far from being the only options. The programmable gate arrays field may be useful in use cases where data paths customized to accelerate workloads, and are used GPU , as seen most often in gaming systems but also in increasing numbers in the world of high performance computers, they are excellent for handling tasks that need a lot of parallel processing.
“Where a good RISC or CISC CPU can have a dozen cores, a large GPU can have a thousand cores,” he said. “And if your system and algorithms can work with parallel processing, GPUs are a very economical and very energy efficient way to get much more for the money.”
Finally, the Tensor processing units , optimized to facilitate automatic learning and AI-based tasks, have applications for obvious applications that depend on that type of functionality.
Storage in fog computing
There is a hierarchy of storage options for fog computing that goes from cheap but slow to fast and expensive . At the first end, that option is storage connected to the network . A NAS offers large volumes of storage, especially in a distributed network, but that means latency times measured in seconds or minutes. The rotating disks might work well for large media libraries or data files, according to Byers, while offering substantially better response times.
Higher in the hierarchy, flash storage, in the form of normal SSDs , provides functionality very similar to that of a rotating tray, with the known consequence of a higher price per GB for much faster access times. That might work better for fast mass storage, though Byers also points out that there are concerns about dropping access speeds after a sufficiently large number of read / write cycles.
“After writing to a certain address on the chip more than 2,000 times, it becomes harder to reprogram it, to the point where, eventually, you will get write failures in that sector of the flash drive,” he said. “So, you have to do something called leveling layers in the flash array, so you write all the addresses in the array the same number of times – many times, the flash drive will take care of that.”
Local flash chips, which are not configured in SSD arrays , are a good solution for security keys, tables and log files, and in the most expensive part of the spectrum, main memory. This is more suitable for popular content, in-memory databases, etc.
Network options for fog computing
There is no hierarchy so easy to establish in the variety of network options available to fog architects, which are obviously divided into wired and wireless categories , subdividing the latter into licensed and unlicensed varieties.
Byers offered a less specific orientation on this element, appealing to choose “those that make sense to you”. Wireless technology tends to be low cost and low impact , and is really the only option for a fog deployment that has to relate to mobile devices.
Licensed wireless technology tends to be controlled slightly better , with less potential interference from external sources, but obviously license and / or usage fees will apply.
However, according to Byers, wiring tends to be preferable to wireless, whenever possible, since they are immune to interference and do not use the radio frequency spectrum. “We like wired networks, especially as you get closer to the cloud, because wired networks tend to have more bandwidth and much less operational expenses,” he said.