+ All Categories
Home > Documents > 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan,...

1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan,...

Date post: 03-Jan-2016
Category:
Upload: julius-cain
View: 221 times
Download: 0 times
Share this document with a friend
Popular Tags:
30
1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001, Anchorage, AK, April 2001
Transcript
Page 1: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

1

On the Placement of Web Server Replicas

Lili Qiu, Microsoft ResearchVenkata N. Padmanabhan, Microsoft Research

Geoffrey M. Voelker, UCSD

IEEE INFOCOM’2001, Anchorage, AK, April 2001

Page 2: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

2

Outline Overview Related work Our approach Simulation methodology & results Summary

Page 3: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

3

Motivation Growing interests in Web

server replicas Exponential growth in Web usage Content providers want to offer

better service at lower cost Solution: replication

Forms of Web server replicas Mirror sites Content Distribution Networks

(CDNs) CDN: a network of servers Examples: Akamai, Digital Island

Internetreplica

replica

Clients Content Provider

s

replica

replica

replica

Page 4: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

4

Placement of Web Server Replicas

Problem specification Among a set of N potential sites, pick K sites as

replicas to minimize users’ latency or bandwidth usage

Internet

ClientsContent

Providers

Page 5: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

5

Related Work Placement of Web proxies [LGI+99] Cache location [KRS00] Placement of Internet instrumentation

[JJJ+00]

Page 6: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

6

Our Approach Model Internet as a graph Parameterize the graph using measured inputs

# requests generated from each region Distance between different regions

Map the placement problem onto a graph optimization problem

Assumption: Each client uses a single replica that is closest to it

Solve graph optimization problem Using various approximation algorithms

Page 7: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

7

Minimum K-median Problem

Given a complete graph G=(V,E), d(j), c(i,j)

d(j): # requests c(i,j): distance between node

i and j Latency or hop counts or other metric to be

optimized Find a subset V’ V with |

V’| = K s.t. it minimizes

vV minwV’ d(v)c(v,w) NP-hard problem

2

510

3

8

4 7

3

6

8

25

64 2

Page 8: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

8

Placement Algorithms Tree based algorithm [LGG+99]

Assume the underlying topologies are trees, and model it as a dynamic programming problem

O(N3M2) for choosing M replicas among N potential places

Random Pick the best among several random

assignments Hot spot

Place replicas near the clients that generate the largest load

Page 9: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

9

Placement Algorithms (Cont.)

Greedy algorithm Calculate costs of assigning clients to replicas Select replica with lowest cost Adjust costs based upon assignment, repeat until

done

Super-Optimal algorithm Lagrangian relaxation + subgradient method

Page 10: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

10

Simulation Methodology Network topology

Randomly generated topologies Using GT-ITM Internet topology generator

Real Internet network topology AS level topology obtained using BGP routing data from

a set of seven geographically dispersed BGP peers Web Workload

Real server traces MSNBC, ClarkNet, NASA Kennedy Space Center

Performance Metric Relative performance: costpractical/costsuper-optimal

Page 11: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

11

Simulation Methodology (Cont.)

Simulate a network of N nodes (100 N 3000)

Cluster clients using network aware clustering [KW00]

IP addresses with the same address prefix belong to a cluster

A small number of popular clusters account for most requests

Top 10, 100, 1000, 3000 clusters account for about 24%, 45%, 78%, and 94% of the requests respectively

Pick the top N clusters Map them to different nodes

Page 12: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

12

Simulation Methodology (Cont.) Random trees Random graphs AS-level topologies Sensitivity to the error in the input

Page 13: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

13

Random Tree Topologies

Tree-based algorithm performs well as expected.Greedy algorithm performs equally as well.

Page 14: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

14

Random Graph Topologies

The greedy and hot-spot algorithms out-perform the tree-based algorithm.

Page 15: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

15

Large Random Graph Topologies

The greedy performs the best, and the hot-spot performs nearly as well.

Page 16: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

16

AS-level Internet Topologies

The greedy performs the best, and the hot-spot performs nearly as well.

Page 17: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

17

Effects of Imperfect Knowledge about Input Data

Predicted workload (using moving window average)

Perfect topology information

Within 5% degradation when using predicted workload

Page 18: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

18

Effects of Imperfect Knowledge about Input Data (Cont.) Predicted workload (using moving window

average) Noisy topology information

Perturb the distance between two nodes i and j by up to a factor of 2

Within 15% degradation when using predicted workload and noisy topology information

Page 19: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

19

Summary One of the first experimental studies on placement of

Web server replicas Knowledge about client workload and topology is needed

for provisioning replicas The greedy algorithm performs very well

Within a factor of 1.1 – 1.5 of the super-optimal Insensitive to noise

Stay within a factor of 2 of the super-optimal when the salted error is a factor of 4

The hot spot algorithm performs nearly as well Within a factor of 1.6 – 2 of the super-optimal

Obtaining input data Moving window average for load prediction Using BGP router data to obtain topology information

Page 20: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

20

Conclusion Recommend using the greedy

algorithm for deciding the placement of Web server replicas

Page 21: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

21

Acknowledgement Craig Labovitz Yin Zhang Ravi Kumar

Page 22: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

22

Comments on greedy algorithm performance Worst-case performance: unbounded Bad example

A full homogeneous binary tree with n=2i leaves and n caches

optimal cost = 0 greedy cost = (n-1)*d

However, the worst-case scenario seems unlikely to occur in real and random topologies

0

ddd d

0 0

Page 23: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

23

Simulation Results inRandom Tree Topologies

Page 24: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

24

Random Tree Topologies

Tree-based algorithm performs well as expected.Greedy algorithm performs equally as well.

Page 25: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

25

Random Graph Topologies

The greedy and hot-spot algorithms out-perform the tree-based algorithm.

Page 26: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

26

Large Random Graph Topologies

The greedy performs the best, and the hot-spot performs nearly as well.

Page 27: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

27

AS-level Internet Topologies

The greedy performs the best, and the hot-spot performs nearly as well.

Page 28: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

28

Simulation Results inReal Internet Topologies

Page 29: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

29

Obtaining Input Data Workload

The number of requests generated by popular client clusters

Stable Placement algorithm can use moving window average

for predicting load with negligible impact on performance

Network topology Propagation delay Hop count AS hop count Internet weather map

Page 30: 1 On the Placement of Web Server Replicas Lili Qiu, Microsoft Research Venkata N. Padmanabhan, Microsoft Research Geoffrey M. Voelker, UCSD IEEE INFOCOM’2001,

30

Placement of Web Server Replicas

Goal Placing K replicas to

minimize users’ latency or bandwidth usage

Minimum K-median problem Select K servers to

minimize the sum of assignment costs

NP-hard problem

Internetreplica

replica

replica

replica

replica

ClientsContent

Providers


Recommended