# Analysis of Parallel-Server Systems with Dynamic Affinity Scheduling and Load Balancing - PowerPoint PPT Presentation

Analysis of Parallel-Server Systems with Dynamic Affinity Scheduling and Load Balancing

1 / 25
Analysis of Parallel-Server Systems with Dynamic Affinity Scheduling and Load Balancing

## Analysis of Parallel-Server Systems with Dynamic Affinity Scheduling and Load Balancing

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
##### Presentation Transcript

1. Mark S. Squillante Mathematical Sciences Department April 18, 2004 Analysis of Parallel-Server Systems with Dynamic Affinity Scheduling and Load Balancing

2. Problem Motivation Cache Affinity [SquillanteLazowska90]

3. Problem Motivation Cache Affinity [SquillanteLazowska90] Key Points of Fundamental Tradeoff • Customers can be served on any server of a parallel-server queueing system • Each customer is served most efficiently on one the servers • Load imbalance among queues occurs due to stochastic properties of system

4. General Overview • Optimal Dynamic Threshold Scheduling Policy • Fluid limits • Diffusion limits • Analysis of Dynamic Threshold Scheduling • Consider generalized threshold scheduling policy • Matrix-analytic analysis and fix-point solution, asymptotically exact • Numerical experiments • Optimal settings of dynamic scheduling policy thresholds • Stochastic Derivative-Free Optimization

5. () 1  () P Scheduling Policy

6. () 1  ()  () P Scheduling Model

7. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis State Vector ( i, j, v ): • i: total number of customers waiting or receiving service at the processor of interest • j: number of customers in the process of being migrated to the processor of interest • v: K-bit vector denoting customer type of up to the first K customers at the processor

8. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

9. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

10. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

11. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

12. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

13. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis Final Solution Obtained via Fix-Point Iteration • Initialize (ps, s, pr, r) • Compute stationary probability vector in terms of (ps, s, pr, r) • Compute new values of (ps, s, pr, r) in terms of stationary vector • Goto 2 until differences between iteration values are arbitrarily small

14. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

15. (1-ps) … … …       0,1,0 1,0,1 Tru,1,0 Tru+1,0,1 Tsu,1,0 Tsu+1,0,1 prg r r r … … … pr (1-pr)g       s (1-ps) … … …       0,0,0 1,0,0 Tru,0,0 Tru+1,0,0 Tsu,0,0 Tsu+1,0,0 +r +r +r m m m (1-pr) Mathematical Analysis

16. Mathematical Analysis General K, Pure Sender-Initiated Policy

17. Mathematical Analysis General K

18. Mathematical Analysis General K

19. Numerical Results

20. Numerical Results

21. Numerical Results

22. Numerical Results

23. Numerical Results

24. Stochastic Derivative-Free Optimization • Internal Model • External Model • Trust Region

25. General Overview • Optimal Dynamic Threshold Scheduling Policy • Fluid limits • Diffusion limits • Analysis of Dynamic Threshold Scheduling • Consider generalized threshold scheduling policy • Matrix-analytic analysis and fix-point solution, asymptotically exact • Numerical experiments • Optimal settings of dynamic scheduling policy thresholds • Stochastic Derivative-Free Optimization