The inaugural Fog Computing Conference is taking place this week in San Jose. Carlos Morales helped set the stage for the event, of which Cisco is the Diamond sponsor, yesterday in an afternoon presentation called “A Vision for Fog Software and Application Architecture.”
Fog computing is the middle ground between connected devices and the cloud that leverages intelligence at the edge for more efficient data handling and faster, local decision making. It collects and aggregates data from multiple sensors, and can do simple data analysis at the edge.
Focusing his discussion on edge devices with enough compute, storage, and network resources to provide a decent impression of a cloud service, Morales, technical leader of the core software group architecture team at Cisco, said fog is essentially an extension of the cloud with the same or similar tools, the same deployment model, and top-level management tools that just work.
The cloud and fog have many commonalities, said Morales, and fog can leverage the huge body of work related to key technologies like orchestration and virtualization. Like cloud, fog has multitenancy, said Morales.
Even a public bus requires that, he said, noting multiple departments may need to support such applications as security video, driver analysis, route planning, accident response, telematics, payment systems, and passenger Wi-Fi. But installing and maintaining all the gear for the separate applications is a management and integration nightmare, he said, so you need to find a way to partition. You can do that, he said, by taking one capable compute box, using VMs, then managing connectivity inside software so you don’t have to go in there and patch wires. That makes installation, management, upgrades, expansion, and introducing new services all easier.
While fog and the cloud have much in common, fog is not the cloud, Morales emphasized, pointing out that their topologies are different, and that fog has thing connectivity, and different requirements in terms of latency and jitter.
Legacy data center topologies, he said, involve two or three tiers, and they have dedicated network elements attached by fat pipes. But fog is attached to IoT, he said, and IoT is not so accommodating, so you end up with an “ad hoc best fit topology.”
“Today cloud doesn’t even know about things, so when you spot an application or a virtual machine in the cloud, there’s no way to say that this app needs to be in this fog node,” he said.
Gateways, however, can be used to map things on the thing network to application-visible overlays. There’s also a need for a control app that allows IoT deployments to assess server load and select the load that is least congested. With fog you need to have security, topology awareness, and dynamic scale, said Morales. You need to continuously measure for latency and jitter, and provision fog overlays with policy.
Intelligent networking for cloud is nice to have, he said, but for fog it is critical.
Edited by Maurice Nagle