Presentation is loading. Please wait.

Presentation is loading. Please wait.

VARUN GUPTA Carnegie Mellon University 1 With: Mor Harchol-Balter (CMU)

Similar presentations


Presentation on theme: "VARUN GUPTA Carnegie Mellon University 1 With: Mor Harchol-Balter (CMU)"— Presentation transcript:

1 VARUN GUPTA Carnegie Mellon University 1 With: Mor Harchol-Balter (CMU)

2 2 K homogeneous First-Come-First-Served servers Exponentially-distributed job sizes Load Balancer Knows queue lengths Not job sizes Q: What is the optimal load balancing policy? A: Join-the-Shortest-Queue Q: Why? A: JSQ = Minimize Expected Response time of arrival GOAL: Minimize Mean Response Time E[T] Poisson( λ )

3 3 μ=4μ=4 μ=4μ=4 μ =1 K heterogeneous First-Come-First-Served servers Exponentially-distributed job sizes Load Balancer Knows queue lengths Not job sizes Q: What is the optimal load balancing policy? μ =1 GOAL: Minimize Mean Response Time E[T] Poisson( λ )

4 Smart-JSQ = Join- Shortest-Queue (with smart tie breaks) 4 MER = Minimum Expected Response time μ=4μ=4 μ =1 μ=4μ=4 μ=4μ=4 μ=4μ=4 Q: Which is the better policy? Q: What is the optimal policy?

5 5 Outline Many-servers limit: Simulation Results Effect of K Effect of arrival rate ( λ ) Effect of degree of heterogeneity Light-traffic regimeHeavy-traffic regime Partial characterization of the optimal policy Complete characterization of optimal policies First asymptotic approximations

6 Poisson( λ ) 1 K 1 = α 1 K K 2 = α 2 K

7 K 1 = K/3 K 2 = 2K/3 Poisson( λ ) Q: Performance of MER 1 Case 1: λ < 4K/3 Fast can handle λ Arrivals find at least one fast idle E[T] = 1/4 Case 2: λ > 4K/3 Fast can not handle λ Can not use slow until each fast has 3 jobs !

8 K 1 = K/3 K 2 = 2K/3 Poisson( λ ) Q: Performance of Smart-JSQ 1 Case 1: λ < 4K/3 Fast can handle λ Arrivals find at least one fast idle E[T] = 1/4 Case 2: λ > 4K/3 Use slow as soon as each fast has 1 job !

9 K 1 = K/3 K 2 = 2K/3 Poisson( λ ) Smart-JSQ better than MER! …but any policy which sends to slow when all fast are busy is identical in light-traffic 1

10 HYBRID (smart- JSQ+MER) Smart-JSQ 10 MER μ=4μ=4 μ =1 μ =4 μ =1 Light-traffic HYBRID = Smart-JSQ μ=4μ=4 μ =1 μ =4 μ =1 μ=4μ=4 μ=4μ=4 smart-JSQ when some server idle MER when all busy

11 11 Outline Many-servers limit: Simulation Results Effect of K Effect of arrival rate ( λ ) Effect of degree of heterogeneity Light-traffic regimeHeavy-traffic regime Partial characterization of the optimal policy Complete characterization of optimal policies First asymptotic approximations

12 K 1 = α 1 K K 2 = α 2 K Poisson( λ ) GOAL Analysis of policies for heterogeneous servers Analysis of JSQ for homogeneous server

13 K 1 = α 1 K K 2 = α 2 K Poisson( λ ) GOAL Analysis of policies for heterogeneous servers Analysis of JSQ for homogeneous server

14 14 K Poisson( λ ) Analysis technique: Markov chain for total jobs in system 01K2K+22K+12K2K3K/2 λ λλλλ λ ? = mean departure rate given 3K/2 jobs

15 15 N = 3K/2 Poisson( λ ) K/2 Rate = K/2 Rate = K Departure rate = K–1 (not K) Finding the O(1) fluctuations critical to analysis O(1) idle queues

16 16 N = (1+ γ )K (0 <γ< 1) Poisson( λ ) γKγK (1- γ )K Rate = (1- γ )K Rate = K Departure rate = K–(1- γ )/ γ (not K) Finding the O(1) fluctuations critical to analysis O(1) idle queues

17 17 K Poisson( λ ) Analysis technique: Markov chain for total jobs in system 01K2K+22K+12K2K3K/2 λ λλλλ λ KK K-1 Asymptotically negligible probability mass First closed-form approx for JSQ!

18 K 1 = α 1 K K 2 = α 2 K Poisson( λ ) GOAL Analysis of policies for heterogeneous servers Analysis of JSQ for homogeneous server

19 19 OPT policy maximize departure rate for each N (preemptively) send jobs to slow servers even when they have 1 job and all fast servers have > 1 Smart-JSQ is optimal in many-servers K 1 = α 1 K K 2 = α 2 K Poisson( λ )

20 20 Outline Many-servers limit: Simulation Results Effect of K Effect of arrival rate ( λ ) Effect of degree of heterogeneity Light-traffic regimeHeavy-traffic regime Partial characterization of the optimal policy Complete characterization of optimal policies First asymptotic approximations

21 21 Smart-JSQ

22 22 MER Smart-JSQ

23 23 MER Smart-JSQ HYBRID

24 24 MER Smart-JSQ HYBRID

25 25 Smart-JSQ

26 26 MER Smart-JSQ

27 27 MER Smart-JSQ HYBRID

28 A new many-servers heavy-traffic scaling to analyze load balancing policies First closed-form approx of load balancing heuristics Choosing the right load balancer Few servers, Small load, High heterogeneity HYBRID Many servers, High load, Low heterogeneity smart-JSQ 28

29 29 MER Smart-JSQ HYBRID

30 30 MER Smart-JSQ HYBRID

31 31 MER Smart-JSQ HYBRID


Download ppt "VARUN GUPTA Carnegie Mellon University 1 With: Mor Harchol-Balter (CMU)"

Similar presentations


Ads by Google