Evaluation of Replica Placement and Retrieval Algorithms in Self - - PowerPoint PPT Presentation

▶
evaluation of replica placement and retrieval algorithms
SMART_READER_LITE
LIVE PREVIEW

Evaluation of Replica Placement and Retrieval Algorithms in Self - - PowerPoint PPT Presentation

Aug 09, 2023 583 likes •773 views

Evaluation of Replica Placement and Retrieval Algorithms in Self Organizing CDNs Jan Coppens, Tim Wauters, Filip De Turck, Bart Dhoedt and Piet Demeester IFIP/IEEE International Workshop onSelf-Managed Systems & Services (SELFMAN)

slide-1
SLIDE 1

Evaluation of Replica Placement and Retrieval Algorithms in Self Organizing CDNs

Jan Coppens, Tim Wauters, Filip De Turck, Bart Dhoedt and Piet Demeester

IFIP/IEEE International Workshop onSelf-Managed Systems & Services (SELFMAN)

slide-2
SLIDE 2
  • Dept. of Information Technology - Ghent University

Overview

  • CDN Architecture
  • Replica Placement Algorithms
  • Reference solution by means of an ILP
  • General real-time placement heuristics
  • COCOA heuristic
  • Evaluation of the placement algorithms
  • Complexity and scalability
  • Performance analysis of the RPA algorithms
  • Using Traffic Engineering for load balancing
  • Conclusion and future work

2

slide-3
SLIDE 3
  • Dept. of Information Technology - Ghent University

Content Distribution Networks

  • Replicate and distribute the content to the edges
  • f the network
  • Increase availability and throughput
  • Decrease end-to-end delay and packet loss
  • Focus on the delivery of video streams

3

Central Server Approach Content Distribution Network

Server Origin Server Replica Server Replica Server

slide-4
SLIDE 4
  • Dept. of Information Technology - Ghent University

CDN Architecture

  • Layered architecture for a content

distribution network

  • Consists of multiple functional modules

4

R e p l i c a P l a c e m e n t A l g

  • r

i t h m T i m e I n t e r v a l C

  • n

t e n t P l a c e m e n t D a t a b a s e C

  • n

t e n t D i s t r i b u t i

  • n

M

  • d

u l e C

  • n

t e n t R e t r i e v a l A l g

  • r

i t h m C

  • n

t e n t R e t r i e v a l M

  • d

u l e C D N P l a c e m e n t C

  • n

t e n t P l a c e m e n t C D N N e t w

  • r

k a n d S e r v e r M

  • n

i t

  • r

i n g D i s t r i b u t e d N e t w

  • r

k M

  • n

i t

  • r

s C

  • n

t e n t S e r v e r M

  • n

i t

  • r

s C e n t r a l P r

  • c

e s s i n g U n i t I n t e r f a c e t

  • h

a r d w a r e d e v i c e s

CDN Operation CDN Network Management CDN Hardware

M

  • n

i t

  • r

i n g R e p

  • s

i t

  • r

y C

  • n

t e n t R e m

  • v

a l M

  • n

i t

  • r

e d I n f

  • r

m a t i

  • n

C l i e n t R e q u e s t

Layer 3 Layer 2 Layer 1

C

  • n

f i g u r e a n d I n i t i a l i z e S e r v e r C

  • n

t e n t S e r v e r s H a n d l e C l i e n t R e q u e s t M u l t i c a s t t r e e

  • r

C h a n n e l C r e a t i

  • n

M u l t i c a s t a n d T E M a p T r a f f i c E n g i n e e r e d P a t h T r i g g e r e d E v e n t C

  • n

f i g u r e N e t w

  • r

k I n i t i a t e S e r v e r C

  • n

t e n t L

  • c

a t i

  • n
slide-5
SLIDE 5
  • Dept. of Information Technology - Ghent University

CDN Architecture

  • Layered architecture for a content

distribution network

  • Consists of multiple functional modules

5

R e p l i c a P l a c e m e n t A l g

  • r

i t h m T i m e I n t e r v a l C

  • n

t e n t P l a c e m e n t D a t a b a s e C

  • n

t e n t D i s t r i b u t i

  • n

M

  • d

u l e C

  • n

t e n t R e t r i e v a l A l g

  • r

i t h m C

  • n

t e n t R e t r i e v a l M

  • d

u l e C D N P l a c e m e n t C

  • n

t e n t P l a c e m e n t C D N N e t w

  • r

k a n d S e r v e r M

  • n

i t

  • r

i n g D i s t r i b u t e d N e t w

  • r

k M

  • n

i t

  • r

s C

  • n

t e n t S e r v e r M

  • n

i t

  • r

s C e n t r a l P r

  • c

e s s i n g U n i t I n t e r f a c e t

  • h

a r d w a r e d e v i c e s

CDN Operation CDN Network Management CDN Hardware

M

  • n

i t

  • r

i n g R e p

  • s

i t

  • r

y C

  • n

t e n t R e m

  • v

a l M

  • n

i t

  • r

e d I n f

  • r

m a t i

  • n

C l i e n t R e q u e s t

Layer 3 Layer 2 Layer 1

C

  • n

f i g u r e a n d I n i t i a l i z e S e r v e r C

  • n

t e n t S e r v e r s H a n d l e C l i e n t R e q u e s t M u l t i c a s t t r e e

  • r

C h a n n e l C r e a t i

  • n

M u l t i c a s t a n d T E M a p T r a f f i c E n g i n e e r e d P a t h T r i g g e r e d E v e n t C

  • n

f i g u r e N e t w

  • r

k I n i t i a t e S e r v e r C

  • n

t e n t L

  • c

a t i

  • n

Content Retrieval Content Distribution CDN Monitoring

slide-6
SLIDE 6
  • Dept. of Information Technology - Ghent University

Replica Placement Algorithms

  • Reference solution by means of an ILP
  • Determines the optimal placement of a static

request distribution

  • Evaluation off-line (NP-complete)
  • General real-time RPA heuristics
  • Periodically replaces content in the CDN
  • Evaluation on-line

6

slide-7
SLIDE 7
  • Dept. of Information Technology - Ghent University

Replica Placement Heuristics

  • Random replica placement
  • Popularity algorithms (parallel for each server)
  • Popularity Local (pop-L) - local content popularity
  • Popularity Global (pop-G) - global content popularity
  • Greedy algorithms (sequential execution for

each content position)

  • Greedy Single (gre-S) - cost of retrieving from origin
  • Greedy Global (gre-G) - cost from other servers
  • Greedy All (gre-A) - cost of all streams in CDN

7

slide-8
SLIDE 8
  • Dept. of Information Technology - Ghent University

COCOA RPA

  • COCOA: Co-Operative Cost Optimization

Algorithm

  • Requires the aid of the Content Retrieval module
  • CR module determines the profit of available content
  • r cost of missing content (real-time)
  • The COCOA RPA uses this information to make its

placement decision (through monitoring module)

  • Hybrid algorithm
  • Centralized content retrieval algorithm (also used

with other RPA algorithms, but more intelligent)

  • Distributed replica placement

8

slide-9
SLIDE 9
  • Dept. of Information Technology - Ghent University

RPA Complexity

  • Compare the computational complexity of the

RPA heuristics

  • COCOA has the same complexity as popularity

local, but uses more accurate information

9

RPA Requests Topology Process Complexity Random None None Distributed O(CsS) Pop-L Local None Distributed O(CsSF) Pop-G Global None Hybrid O(CsSF) Gre-S Local Origin Distributed O(CsSF) Gre-G All Entire Centralized O(CsS3F) Gre-A All Entire Centralized O(CsS4F) COCOA CR Module None Hybrid O(CsSF)

slide-10
SLIDE 10
  • Dept. of Information Technology - Ghent University

RPA Complexity

  • Compare the computational complexity of the

RPA heuristics

  • COCOA has the same complexity as popularity

local, but uses more accurate information

10

RPA Requests Topology Process Complexity Random None None Distributed O(CsS) Pop-L Local None Distributed O(CsSF) Pop-G Global None Hybrid O(CsSF) Gre-S Local Origin Distributed O(CsSF) Gre-G All Entire Centralized O(CsS3F) Gre-A All Entire Centralized O(CsS4F) COCOA CR Module None Hybrid O(CsSF) O(Cs) O(CsF) O(CsF) O(CsF) O(CsF)

Parallelism

slide-11
SLIDE 11
  • Dept. of Information Technology - Ghent University

Performance Analysis

  • Overhead of the average load of the algorithms

to the ILP solution

  • COCOA is better than pop-L and close to the

gre-G placement

11

Oslo Stockholm Copenhagen Amsterdam Dublin London Brussels Paris Madrid Zurich Milan Berlin Athens Budapest Vienna Prague Warsaw Munich Rome Hamburg Barcelona Bordeaux Lyon Frankfurt Glasgow Belgrade Strasbourg Zagreb

25% 50% 75% 100% 125% 150% 175% 200% 0% 50% 100% 150% 200% 250%

Replication Factor Overhead of AVG load to ILP

Random Popularity Local Popularity Global Greedy Single COCOA Greedy Global Greedy All

slide-12
SLIDE 12
  • Dept. of Information Technology - Ghent University

Performance Analysis

  • Overhead of the average load of the algorithms

to the ILP solution

  • COCOA is better than pop-L and close to the

gre-G placement

12

Oslo Stockholm Copenhagen Amsterdam Dublin London Brussels Paris Madrid Zurich Milan Berlin Athens Budapest Vienna Prague Warsaw Munich Rome Hamburg Barcelona Bordeaux Lyon Frankfurt Glasgow Belgrade Strasbourg Zagreb

25% 50% 75% 100% 125% 150% 175% 200% 0% 50% 100% 150% 200% 250%

Replication Factor Overhead of AVG load to ILP

Random Popularity Local Popularity Global Greedy Single COCOA Greedy Global Greedy All

slide-13
SLIDE 13
  • Dept. of Information Technology - Ghent University

Traffic Engineering for Load Balancing

  • Because of asymmetric topology and request

distribution, the load is distributed unevenly over the network

  • Can cause congestion in certain parts of the

network (e.g. during a flash crowd)

  • Traffic Engineering can be used to spread the

flows over the entire CDN

  • Proactive in order to off-load core edges
  • Reactive in order to route flows round congested

bottlenecks

13

slide-14
SLIDE 14
  • Dept. of Information Technology - Ghent University

200 400 600 800 1000 40 80 120 160

Time (minutes) Load on core edges Average load: 42 Standard deviation: 18.4

Proactive Traffic Engineering (1)

14

slide-15
SLIDE 15
  • Dept. of Information Technology - Ghent University

200 400 600 800 1000 40 80 120 160

Time (minutes) Load on core edges Average load: 42 Standard deviation: 18.4

Proactive Traffic Engineering (1)

15

200 400 600 800 1000 40 80 120 160

Time (minutes) Load on core edges Average load: 57.5 Standard deviation: 5.5

slide-16
SLIDE 16
  • Dept. of Information Technology - Ghent University

200 400 600 800 1000 40 80 120 160

Time (minutes) Load on core edges Average load: 42 Standard deviation: 18.4

Proactive Traffic Engineering (1)

16

200 400 600 800 1000 40 80 120 160

Time (minutes) Load on core edges Average load: 57.5 Standard deviation: 5.5

36.9%↑ 70.1%↓

slide-17
SLIDE 17
  • Dept. of Information Technology - Ghent University

Proactive Traffic Engineering (2)

  • Using Traffic Engineering the average load

increases, but the standard deviation drops

  • The load on the core edges decreases:

17

0% 9% 18% 26% 35% 1 2 3 4 5 6 7 8 9 10 11 12

Drop in edge load Most popular edges where congestion occurs

slide-18
SLIDE 18
  • Dept. of Information Technology - Ghent University

Conclusion and Future Work

  • Novel hybrid CDN architecture and COCOA RPA
  • COCOA placement algorithm
  • Nearly as scalable as popularity local
  • Close to the performance of greedy global
  • Traffic engineering used to off-load the core edges
  • Influence of replacements on the load of the

network?

  • Frequency of RPA executions?
  • How can we make content retrieval distributed?

18