Download presentation
Presentation is loading. Please wait.
Published byWinston Adcox Modified over 9 years ago
1
Bilal Gonen, Murat Yuksel, Sushil Louis University of Nevada, Reno
2
Motivation & Problem definition: Intra- domain traffic engineering Black-box problem and search algorithms. PTAS: A hybrid search algorithm Experimental results Future work
3
Online configuration of large-scale systems such as networks require parameter optimization (e.g. setting link weights) to be done within a limited amount of time. This time limit is even more pressing when configuration is needed as a recovery response to a failure (link failures) in the system.
5
Routers flood information to learn topology Determine “next hop” to reach other routers… Compute shortest paths based on link weights Link weights configured by network operator 5 5 5 5 5 5 5 5 5 5 J. Rexford et al., http://www.cs.princeton.edu/courses/archive/spr06/cos461/http://www.cs.princeton.edu/courses/archive/spr06/cos461/
6
How to set the weights? Inversely proportional to link capacity? Proportional to propagation delay? Network-wide optimization based on traffic? 5 5 5 5 5 5 5 5 5 5 J. Rexford et al., http://www.cs.princeton.edu/courses/archive/spr06/cos461/http://www.cs.princeton.edu/courses/archive/spr06/cos461/
7
Empirical way: Network administrator experience Trial and error error-prone, not scalable 5 5 5 5 5 5 5 5 5 5 J. Rexford et al., http://www.cs.princeton.edu/courses/archive/spr06/cos461/http://www.cs.princeton.edu/courses/archive/spr06/cos461/
8
5 5 5 5 10 5 5 5 5 5 5 congestion 20
10
Which algorithm is better? NFL Theorem(Wolpert, 1997): No matter what perform metric is used, the average performance of any search algorithm will be the same over all possible problems. General-purpose universal algorithm is impossible
11
Exploration: global phase, examine overall features, supply effectiveness Exploitation: local phase, examine microscopic features, supply efficiency
12
Exploration techniques: Random sampling Random walk Genetic Algorithm Exploitation techniques: Downhill simplex Hillclimbing Simulated Annealing Hybrid Recursive Random Search (RRS), T. Ye et al. ToN 2009
13
Exploration techniques: Random sampling Random walk Genetic Algorithm Exploitation techniques: Downhill simplex Hillclimbing Simulated Annealing Hybrid Recursive Random Search (RRS), T. Ye et al. ToN 2009
14
An algorithm may be good at one class of problems, but its performance will suffer in the other problems Key Question: How to design an evolutionary hybrid search algorithm? search for the best search Roulette wheel: Punish the bad algorithms and reward the good ones trans-algorithmic Transfer the best-so-far among the algorithms
16
The best known strategy to select among slot machines for investment! Viewing each algorithm as a slot machine!
17
Total Budget = 1500 300 Round budget = 300 Algorithm-1 Round-1 budget=100 Algorithm-2Algorithm-3 budget=100 Winner Algorithm-1 Round-2 budget=110 Algorithm-2Algorithm-3 budget=98 budget=92 Winner Algorithm-1 Round-3 budget=106 Algorithm-2Algorithm-3 budget=90 budget=104 Winner Algorithm-1 Round-4 budget=120 Algorithm-2Algorithm-3 budget=92 budget=88 Winner Algorithm-1 Round-5 budget=110 Algorithm-2Algorithm-3 budget=102 budget=88 Winner
18
We used several benchmark objective functions to model the black-box system: Square Sum function Rastrigin function Griewangk’s Function Axis parallel hyper-ellipsoid function Rotated hyper-ellipsoid function Ackley’s Path function. PTAS outperforms all the other three algorithms for most of the objective functions
19
Square Sum function: f1(x)=sum(x(i)^2), i=1:n
20
Griewangk's function f(x)=sum(x(i)^2/4000)-prod(cos(x(i)/sqrt(i)))+1, i=1:n
21
PTAS’ benefits are more pronounced when objective function changes SquareSum -> Rastrigin -> SquareSum
22
PTAS’ benefits are more pronounced when objective function changes Griewangks -> Rastrigin -> Griewangks
23
Simulation Setup: Ns-2 Exodus topology from Rocketfuel # of flows: 90 # of nodes: 22 # of links: 37 Confidence interval: 80% (We repeated the optimization process 30 times to gain confidence.) Optimization objective: minimize the overall packet drop rate. Thus, maximize aggregate network throughput
24
PTAS NS-2 simulator Aggregated Througput Change link weights
25
6% 15% 33%
26
PTAS framework is applicable to the various network configuration problems, e.g.: Random Early Detection (RED) queue management algorithm BGP inter-domain traffic engineering
27
Thank you…
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.