Computing delays etc for WFQ
Assume all connections are leaky-bucket constrained: connections 1,2,..., N; connection i is (si, ri) -constrained
Compute delay for GPS, and use the relationship fi,WFQk < fi,GPSk + Lmax / r
Theorem: Connection i achieves maximum delay and backlog under GPS if each connection begins generating its maximum possible traffic at the same instant
- Maximum traffic: A burst of size si arrives at time 0; thereafter, traffic arrives at a rate ri
- aka “greedy traffic arrival”; “critical instance” of arrivals
Intuitively, makes sense
- “stressing” the system the most...
Proof: [Parekh & Gallager ’93]; similar to [Liu & Layland ’73]