Telefonica: “I feel one of many killer apps of community programmability may be ML”
Earlier than diving into the position of a programmable knowledge airplane in supporting telco AI functions, Telefonica Analysis’s Senior Analysis Scientist Eduard Marin Fabregas supplied a little bit of a historical past lesson to attendees of the Telco AI Discussion board 2.0, out there on demand right here. First issues first, he famous the standard community composition of a administration airplane to watch and configure units remotely, a management airplane, together with protocols to populate forwarding tables, and an information airplane made up of all of the units used to ahead packets.
This setup, Fabregas stated, was designed “with resiliency in thoughts, which was nice within the early days. But it surely additionally got here with some issues. One of many issues…is that the info airplane and the management airplane are tightly coupled collectively on the save system.” {Hardware}-centric routers lacked programmability, supported a set set of functionalities, and the ASICs they’re constructed with led to value and complexity. “This prevents us from innovating,” he stated. “And, in fact, this has had some influence on many functions that we may develop within the networks. AI is one in all them.”
The rise of software-defined (SDN) controllers, and different applied sciences like community features virtualization, container administration and eBPF, have enhanced the flexibility to watch packets and knowledge flows, and different community telemetry, and have in any other case elevated programmability. Particular to SDN, Fabregas stated, open, commonplace interfaces and disaggregation have enabled a bigger ecosystem to ship extra innovation. “And this has modified rather a lot the best way we are able to do ML within the community.”
He continued: “Now what we are able to do is we are able to do characteristic assortment on the info airplane, so on the router itself.” Nevertheless, inference can’t be accomplished within the routers. As a substitute, operators configure the info airplane to pattern particular knowledge factors typically both periodically or as triggered by outlined occasions; that info can then be forwarded to the management airplane. “The issue is we are able to’t actually do per packet inference at dwell pace…Nonetheless this isn’t best. However this isn’t the top of the story.”
He stated that utilizing programmable knowledge planes for ML duties is “one of many largest improvements that we’ve had in a few years.” Utilizing the P4 programming language, CSPs can primarily set routers as much as conduct specified operations at Terabit pace for real-time resolution making and improved community visibility. “We will determine issues on the community in actual time on the routers themselves…We will extract many extra insights from the community.” As a result of routing and switching infrastructure sits between consumer units and the core community, “It could actually play an necessary position in lots of extra of the purposeful elements behind ML,” Fabregast stated.
Fabregast went on to explain a protocol-independent change structure (PISA) consisting of a parser, programmable pipeline and de-parser (see above picture). Basically, he stated, PISA permits for a number of phases the place ML inference may be utilized, albeit with challenges round reminiscence. “These units supply enormous alternatives for parallelization, so operations that don’t rely upon one another may very well be positioned on the identical stage.” Including ML inferencing within the knowledge airplane permits for extra complicated characteristic extraction and customization.
Fabregast gave the instance of anomaly detection historically accomplished within the router by sampling packets, and doubtlessly forwarding these packets to a different equipment working an ML mannequin to determine, as an illustration, if a packet is malicious or not, then sending it again to the router the place a selected coverage may be utilized. Placing that ML mannequin immediately on the change may assist anomaly detection of each single packet with out affecting throughput. He described this use of a programmable knowledge airplane as “a primary line of protection. You could possibly consider this as deploying a comparatively light-weight, easy ML mannequin maybe simply to detect assaults.”
Huge image, Fabregast stated, “The long run is self-driven autonomous networks—networks that may make choices based mostly on knowledge.” Adopting closed-loop automation for monitoring, evaluation and motion will result in adaptive, resilient networks. Developments in {hardware}, unified requirements and APIs to share info between switches, and additional mannequin growth utilizing artificial and augmented knowledge, will drive additional programmability. “I feel one of many killer apps of community programmability may be ML,” he stated.
