Traditional congestion recovery methods
In traditional networks, addressing congestion is a slow and reactive process.
The control plane must detect reduced bandwidth, update routing tables, and notify all switches in the network. This chain of actions often takes significant time, leading to persistent bottlenecks and degraded performance.
While sufficient for legacy networks, these methods are too slow and inefficient to handle the demands of modern AI fabric environments, where rapid congestion isolation and recovery are critical.
Intelligent local congestion recovery
Cisco Silicon One introduces autonomous local congestion recovery, significantly reducing time-to-resolution.
When a local link becomes congested, the device's MAC layer notifies the Intelligent Collective Networking (ICN) block. The ICN dynamically adjusts traffic distribution to reduce pressure on the congested link without requiring additional intervention.
This seamless recovery supports both dynamically load-balanced flows and static ECMP flows, ensuring optimal performance and minimal disruption.
Switch congestion notification packet (CNP)
Cisco Silicon One ensures fast convergence in extreme congestion scenarios through the Switch Congestion Notification Packet (CNP) mechanism.
When a packet encounters a severely congested path, the packet is forwarded to the destination and also the truncated packet is sent back to the sender.
The sender reacts to this notification by re-pathing the flow, such as by adjusting the flow's entropy value to utilize less congested paths.
This mechanism enables rapid traffic rerouting, reducing bottlenecks and maintaining efficient network operations.
Fabric routing management (FRM)
In traditional networks, the control plane is responsible for detecting congestion and notifying peer devices, a process that can take hundreds of milliseconds.
Cisco Silicon One's hardware-accelerated Fabric Routing Management (FRM) protocol eliminates these delays by monitoring congestion directly in hardware and propagating updates across the network.
FRM dynamically adjusts traffic flows to avoid congested paths, redistributing traffic to less utilized links. This ensures near-instantaneous recovery, reducing convergence time to just a few microseconds and maintaining seamless network performance.
Intelligent bandwidth adjustment
In congestion scenarios, where the bandwidth of a link is significantly reduced but the connection remains operational, Cisco Silicon One ensures intelligent traffic distribution.
The hardware-accelerated routing protocol continuously monitors available bandwidth and dynamically adjusts traffic flows to match the reduced capacity.
Traffic is redistributed across higher-bandwidth links, ensuring optimal utilization of the network and minimizing the impact of congestion on overall performance.