XKL has expanded the data center interconnect (DCI) capabilities of its MediaLight optical transport family for distributed AI and HPC environments, targeting low, deterministic latency between campus, metro, and regional compute clusters. The company says its MediaLight Xponder and Muxponder 1RU systems deliver sub-5 µs optical transport to help reduce latency and keep GPUs from sitting idle during synchronization and large dataset movement.
The update centers on running traffic at Layer 1 over DWDM, with XKL positioning MediaLight as an alternative to packet-layer, multi-hop routing paths that can introduce jitter and queueing delay. For operators stretching training or inference across multiple sites, that’s a practical design trade: fewer packet-layer touch points can mean more predictable behavior, but it also shifts more of the “intelligence” of traffic engineering to how you design the optical layer and the surrounding network.
On the interface side, the MediaLight 1RU platforms support soft-configurable 100G QSFP28 and 400G QSFP-DD transport. XKL also describes “bit-transparent” physical transport intended to carry uncompressed traffic over dark fiber or third-party open line systems (OLS).
For operations teams, XKL says MediaLight includes a standard CLI and an open MIB architecture to integrate with existing monitoring tools, and it emphasizes rapid service turn-up without proprietary management software. That matters in mixed environments where optical gear needs to slot into established NMS workflows, rather than becoming yet another standalone system to babysit.
“Distributed AI workloads cannot afford the latency penalties and packet queueing inherent in traditional routing architectures,” said Casey Inman, Director of Sales & Marketing at XKL. “MediaLight solves this directly at Layer 1.”
XKL says the MediaLight product family is shipping for AI interconnect and enterprise deployments, with product information available on the MediaLight Muxponder and MediaLight Xponder pages.
Source: XKL
