You are redesigning a high-speed transit network due to congestion-related issues. Which congestion avoidance mechanism can you apply to the existing network?

NBAR (Network-Based Application Recognition) is a deep-packet inspection classification engine used to identify application traffic for policy application and has no congestion avoidance function.

FIFO queuing is the default, non-intelligent scheduling method that uses tail-drop when full, which worsens congestion by triggering TCP global synchronization rather than avoiding it.

WRED (Weighted Random Early Detection) monitors queue depth and begins randomly dropping packets - weighted by DSCP or IP precedence - before the queue is full, signaling individual TCP senders to back off gradually. This avoids TCP global synchronization where all flows simultaneously reduce their window size and restart, which is the root cause of the throughput oscillation seen in congested high-speed transit networks.

Rate-limiting (traffic policing) enforces a configured bandwidth ceiling by dropping or remarking excess traffic at ingress but does not manage queue depth or provide active congestion avoidance inside the network.

Policy-Based Routing (PBR) redirects packets to alternate next-hops based on match criteria for traffic engineering purposes and has no mechanism for congestion avoidance.