Edge computing has become a hot topic within networking, but it lacks a precise definition. This week, SDxCentral’s research department took some initial jabs at defining “the edge.” And we’re not talking about the guitarist for U2, here.
One of the difficulties in defining edge computing is determining where the edge is.
“Edge computing is typically located at the access point, one hop away from the user,” according to SDxCentral’s newest report “Innovations in Edge Computing and MEC.”
This may be the radio access network (RAN) for wireless networks, the radio network controller for WiFi, the cable modem termination system (CMTS) for cable, or the optical line terminal (OLT) for a fiber connection. Or it could be an Internet of Things (IoT) gateway or a micro data center in a wide range of locations.
As part of the research, SDxCentral conducted a survey in August 2017 of a select subset of 119 readers familiar with the nascent topic of edge computing. Respondents were asked which hardware platform edge infrastructure would most likely be built on, and they could reply more than once.
Eighty percent of respondents said it will likely be specialized edge servers that can accommodate lower power and cooling needs. While 65 percent said it will be converged edge pods (fully-integrated hyperconverged multinode platforms); and 52 percent think the infrastructure could end up being built on standard data center servers.
Open Compute Project
The Open Compute Project (OCP) is working on open hardware specifications for data centers. But these same specifications could become relevant for edge computing locations as well.
In 2016, OCP accepted the Carrier Grade (CG)-OpenRack-19 specification, which Radisys developed for data center rack architecture. The design is an intersection between Facebook’s OpenRack specification and the legacy 19-inch EIA-310 rack architecture, according to Bill Carter, chief technology officer with OCP.
And in early 2017, Adlink contributed the OpenSled specification for CG-OpenRack-19 — a compatible, open CPU sled.
“We’re not focused on the edge,” said Carter. “We’re really focused at the data center. But when it makes sense, we look at the edge. From the OCP perspective, we’re trying to stay focused on hardware and cloud computing in the data center. We do sometimes get dragged outside the walls of the data center.”
Building off the CG-OpenRack-19 spec and the OpenSled spec, Adlink has developed its own Open Compute Carrier-grade Edge Reference Architecture (OCCERA). OCCERA is a product line of Adlink that is focused on edge computing. It is not open-sourced through OCP.
OCCERA decouples compute, storage, and IO into different functional modules. Based on the requirements of a specific application, these modules are chosen in the necessary proportions, and then combined into a specific hardware entity.