Systematic comparison of scheduler designs for L4S and classic traffic

Determine how strict-priority scheduling compares systematically with weighted schedulers such as weighted round robin and schedulers that guarantee the classic queue a minimum service share, including the resulting trade-offs among L4S isolation, L4S capacity, and completion times for short classic transfers.

Background

The proposed deployment strategy uses strict-priority queuing with a rate cap to isolate L4S traffic at shared sender-access and peering bottlenecks. The experiments show that this configuration protects L4S latency and performance but can substantially increase the flow-completion time of short CUBIC transfers, which are particularly sensitive because they spend much of their lifetime in slow start.

The paper identifies alternative scheduler designs, including weighted round robin and schemes guaranteeing the classic queue a minimum service share. These alternatives could reduce the impact on short classic transfers, but potentially at the cost of weaker L4S isolation or lower L4S capacity. The paper leaves a systematic comparison of these scheduler choices unresolved.