SAN JOSE, Calif. — An open-source group aims to release by April 30 code for carrier edge networks initially driven by AT&T and a team of its vendors. If successful, the Akraino Edge Stack software will someday power “cookie-cutter” deployments of “thousands and tens of thousands … of baby clouds,” said an AT&T executive.
“We’re going to disaggregate the [cellular] radio-access network so it won’t be one big box but many distributed systems using open interfaces,” said Mazin Gilbert, vice president of AT&T Labs, in an interview with EE Times.
The move is “a key ingredient to support our 5G story — to gain speed, minimize latency, deliver security, and minimize dataflow to the core that you need to do more smart processing at the edge,” Gilbert said.
AT&T is one of many carriers gearing up for edge clouds to deliver new services at lower costs. The business rationale is that “our traffic is exponentially increasing, but our revenue is not, so we need options with interoperability,” he said.
The carrier has already delivered to Akraino three or four software programs for high-performance computing, traffic routing, and other uses. They assume “restrictions in networking bandwidth, power, and space” for the edge clouds, which AT&T previously described as varying from a rack of systems in a closet to a small data center.
For efficiency, the designs “have to be a cookie cutter,” enabling high levels of automation because AT&T’s “largest cost is [systems] integration and testing.” The code from Akraino, a project under the Linux Foundation, will ultimately be integrated into “a blueprint of software and hardware Legos,” Gilbert said.
The edge clouds will be based on off-the shelf hardware that includes Arm, x86, and GPU servers. Akraino plans to acquire less than $250,000 in systems for its initial testing, some of them donated.
The deployments will be gradual. “We’re testing a bunch of these technologies for the first time … We start with a crawl and put out a few sites,” he said, noting that AT&T launched 5G in 12 markets late last year.
“We learn from each one — the world has never disaggregated the RAN before,” he added.