Estimate matched traffic, ACE evaluations and CPU cost when PBR runs in hardware versus in software.
In hardware, PBR is a TCAM lookup that costs essentially nothing per packet. Once the policy exceeds TCAM capacity or uses an unsupported match, packets are punted to the CPU and cost is linear in rate — which is what a CPU load above 100 percent in this calculator is telling you. A policy that works in the lab and melts in production is almost always one that fell out of TCAM, and the software ceiling tells you the packet rate at which that happens rather than after the outage.
Policy-Based Routing
Matched packets = rate × match percentage; in software each one costs the per-packet processing time, so CPU load = matched pps × µs per packet ÷ 1,000,000.
Punt-path capacity
software forwarding ceiling = 1,000,000 ÷ µs per matched packet.
Matched packets = rate × match percentage; in software each one costs the per-packet processing time, so CPU load = matched pps × µs per packet ÷ 1,000,000. In hardware, PBR is a TCAM lookup that costs essentially nothing per packet. Once the policy exceeds TCAM capacity or uses an unsupported match, packets are punted to the CPU and cost is linear in rate — which is what a CPU load above 100 percent in this calculator is telling you.
A policy that works in the lab and melts in production is almost always one that fell out of TCAM, and the software ceiling tells you the packet rate at which that happens rather than after the outage.
This calculator takes 6 inputs: Total packet rate, Traffic matching the policy, Route-map clauses, Average ACEs evaluated per packet, Matched in hardware (TCAM), Software cost per matched packet. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
Check the TCAM or ACL resource utilisation counters and the CPU punt rate on the platform. Hardware PBR shows no growth in punted packets under load; if the punt counter tracks the matched rate, the policy is running in software.