Summary: Competition Yields Efficiency in
Load Balancing Games
Jonatha Anselmi (BCAM)
Urtzi Ayesta (LAAS-CNRS)
Adam Wierman (Caltech)
Load balancing game
Each user seeks to minimize its own mean sojourn time
Load balancing game (cont.)
Well studied model: [Bell&Stidham83],[Haviv &
Existence, Uniqueness of equilibria [BS83].
- [RT02] Linear latency functions PoA=4/3
- [HR07] M/M/1-type latency, PoA=N.
- [AAP11] M/M/1 with server dependent costs, unbounded
- [CMW09] SRPT-type, PoA unbounded with 2 servers