1: \begin{abstract}

2: We study a dynamic scheduling problem for a multi-class queueing network

3: with a large pool of statistically identical servers. The arrival processes

4: are Poisson, and service times and patience times are assumed to be

5: exponentially

6: distributed and class dependent. The optimization criterion is the

7: expected long

8: time average (ergodic) of a general (nonlinear) running cost function

9: of the

10: queue lengths. We consider this control problem in the Halfin--Whitt

11: (QED) regime, that is,

12: the number of servers $n$ and the total offered load~$\mathbf{r}$

13: scale like

14: $n\approx\mathbf{r}+\hat{\rho} \sqrt{\mathbf{r}}$ for some constant

15: $\hat\rho$. This problem was proposed in [\textit{Ann. Appl. Probab.} \textbf{14}

16: (2004) 1084--1134, Section~5.2].

17: 

18: The optimal solution of this control problem can be approximated by

19: that of the

20: corresponding ergodic diffusion control problem in the limit. We

21: introduce a broad

22: class of ergodic control problems for controlled diffusions, which

23: includes a large

24: class of queueing models in the diffusion approximation, and establish

25: a complete

26: characterization of optimality via the study of the associated HJB equation.

27: We also prove the asymptotic convergence of the values for the multi-class

28: queueing control problem to the value of the associated ergodic diffusion

29: control problem. The proof relies on an approximation method by

30: \emph{spatial truncation} for the ergodic control of diffusion

31: processes, where

32: the Markov policies follow a fixed priority policy outside a fixed

33: compact set.

34: \end{abstract}