AI Fabric Validation at Scale - VIAVI at MWC Barcelona 2026

Added: 2026-05-15

[00:00:03]Welcome at the MWC in Barcelona today.

[00:00:05]I'm here to talk to you about our AI data fabric solution from VIAVI Solutions. It helps our customers to validate their AI fabric data centers before they put it into production. To do this, we are using our B3 test center appliance, which allows us to emulate GPUs and generate real-life RoCE v2 data center traffic protocol together with a com- communication library CCLs that allow test engineers to really test and validate the AI data traffic going through the network before putting it into a production network. For this demo, we are working together with our partner Nokia that provided our us the switch where our data traffic will pass through. And to connect both devices together with each other, we're using the Amphenol 400 gig QDD DAC cables.

[00:00:58]What we are doing now in our demo to show this is we are emulating the GPUs.

[00:01:03]They are then sending traffic through the network through the Nokia device under test. And by that, we can calculate and show the job completion time. This job completion time is important to be able to tell the customers how efficiently the GPUs are being used and to see how long a job would take to complete. This on his hand gives an indication how long an end user will need to wait until on the prompt he put into the AI engine is provided. In the demo itself, we start with an ideal situation where there is no congestion at all, but then we change the configuration on the Nokia switch to introduce congestion on purpose.

[00:01:48]Congestion meaning that there is too much bandwidth, too much traffic going over the the provided uh, connection.

[00:01:56]Once that happens, the Nokia switch will start to introduce and send ECN congestion notification messages to our VIAVI test center appliance.

[00:02:06]When those arrive at the test center appliance, we will react on it by throttling back, by lowering the bandwidth, thus increasing the job completion time, which in fact means that there will be a less efficient use of the GPUs, which is a costly matter as well, because they are the most expensive part of the entire AI data center solution. What we can show as well with that one is the next to all the accuracy for the testing, we also provide a capability to to use the VIAVI solution to test normal routing, switching, and Ethernet testing as well, which creates an advantage and a better return on investment. By using our simulated GPUs as well, it's not just the testing that makes it easier, but it takes away the needs to create and build a lot of complete GPU farm, which is extremely expensive and consumes a lot of power. By using the simulated GPUs, you can reduce the power consumption by roughly 70%. So, it makes it easier, more repeatable, and better to to test all the the net the AI fabric before it's put into production.

[00:03:20]To configure the test on the VIAVI test center solution, we are using our web UI, which is completely web-based, so it doesn't require any installation on any of the PC and can be accessed from anywhere. It is also wizard-based, so it's very easy to configure the complex uh, environment where you have to emulate the GPUs, the Rocky V2 traffic, and the CCL traffic patterns. To finalize, I want to thank you for your attention, and if you want to get more information on all our AI infrastructure solutions, feel free to visit our website at www.fiammisolutions.com and reach out to anybody on of us to get more information and details on any of our solutions.