Spirent sees an increased focus on AI testing across telecom, driven by two ‘mega-trends’
The booming growth in AI is impacting testing considerations in a number of ways. In a recent RCR Wireless News webinar, Stephen Douglas, Spirent Communications’ head of market strategy, outlined some of the factors in a rapidly changing landscape.
Douglas framed the changes in terms of two mega-trends.
The first is networks being enhanced with artificial intelligence tools. “We’re seeing at the moment, quite a lot of vendors of network equipment starting to bring in AI capabilities into their solutions—whether that’s in switches and routers, whether it’s in the radio equipment, firewalls, gateways, et cetera, even into the core network itself,” Douglas explained. It means that AI is being used, for example, to support things like dynamic policy configurations and that telcos are looking at how it can be embedded in the Radio Access Network for load balancing, increasing energy efficiency, mobility optimization and so on.
“We’re seeing sort of a raft of AI coming into the networks in this way,” Douglas added.
How does that impact testing? The efficacy of that AI has to be confirmed, both before it is deployed and once it is active in the network. Douglas summarized some of the questions that testing of AI network tools have to answer: First off, is the AI working? Are the results as good or better than older, non-AI-based systems? Is it delivering benefits, or is the AI introducing new risks?
The second mega-trend is networks being built to support the use of AI and its demands for compute power, bandwidth, latency and so on. Data centers are already feeling these effects and needing changes in their design and architecture in order to support GPU clusters in terms of power, but Douglas said that the traffic behavior and performance demands within data centers also change with AI—and those changes are also impacting the wider wireline and wireless networks the amount of AI traffic goes up. (ChatGPT alone, it should be noted, this week hit the milestone of having 400 million active weekly users.)
As service providers look to optimize and upgrade their networks to support AI traffic, they also have to test for the parameters that demonstrate whether their networks can provide the lower latency, losslessness, throughput and specific performance characteristics that AI workloads demand.
Douglas identified three capabilities that are enabling AI testing, both in terms of the use of AI in the network, and in networks’ ability to support the adoption and AI. Those are:
-Digital twins, or using emulated replicas of networks as playgrounds to test out AI.
-Synthetic test data, in terms of using realistic but emulated data to either fuel AI training, or test the systems.
-Continuous and active testing—not only in lab environments, Douglas pointed out, but within the live network.
He offered up a couple of examples: When architecting and building out new data center fabrics for AI support, Douglas said, “up until this point in time, the only way to test that [has been] using real XPUs, real GPUs to actually stimulate load. And that’s very expensive, very time consuming and not very realistic.” But now, a digital twin can be used to generate various types of traffic and behavior without the overhead and costs of having to use actual compute XPUs.
Douglas also said that digital twins are being used for security testing, to look at the impacts of realistic attack traffic and impairments, as well as to check that AI-equipped firewalls are actually treating the traffic the way that they are supposed to.
For more insights on AI testing and trends, watch the full webinar on demand, featuring Spirent Communications, Viavi Solutions and DriveNets, and read more of RCR Wireless News’ AI coverage here.
