This article discusses the issue around higher latencies that were observed in a network environment that had varying circuit/link speeds and strict service-level agreements around the minimum/average and maximum latencies. Testing revealed there were increased/higher latencies at lower circuit speeds.

The network environment started off with 40G connections and consistently lowered the port speeds to 20G, 10G, 1G, 0.1G and then to 500Mbps with multicast as well as unicast traffic.  Increased latency was reported  at lower port speeds. Testing was done across JUNOS 15.1F5 and 17.1 code, and the same behavior was seen on MPC7E cards.

This was also tested and replicated in the JTAC lab. The following is the recorded latency numbers for unicast and multicast traffic at different circuit speeds. All values are in microseconds and involve a 3-member network with MPLS in the core and RSVP signaling with aggregated links used ever:

The latency is a function of how busy the PFE gets at varying link speeds for transit traffic. So at higher circuit speeds, the time interval between the packets is less and the PFE is constantly busy processing traffic. Hence, the observed latency is low. In contrast, at lower speeds, because the traffic rate is low, the time interval between packets is really high which causes the PFE to remain idle and get busy only when it receives each packet. As a result, the PFE takes a longer time to process each packet and send it out, which constitutes towards the higher latency. 

This is behavior is by design.