Cost-Efficient Hosting and Load Balancing of Massively Multiplayer Online Games
5K - views

Cost-Efficient Hosting and Load Balancing of Massively Multiplayer Online Games

Similar presentations


Download Presentation

Cost-Efficient Hosting and Load Balancing of Massively Multiplayer Online Games




Download Presentation - The PPT/PDF document "Cost-Efficient Hosting and Load Balancin..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.



Presentation on theme: "Cost-Efficient Hosting and Load Balancing of Massively Multiplayer Online Games"— Presentation transcript:

Slide1

Cost-Efficient Hosting and Load Balancing of Massively Multiplayer Online Games

Vlad Nae, Radu Prodan, Thomas FahringerInstitute of Computer ScienceUniversity of Innsbruck

10/25/2010

Slide2

MMOG Subscriptions (last 10 years)

1. Introduction

24 million subscribers

Source: http://www.mmogchart.com

10/25/2010

Slide3

Entertainment Industries’ Sizes

Entertainment Software Association (ESA)Size: 7 billion $Dynamic: +300% in the last 10 yearsMotion Picture Association of America (MPAA)Size: 8.99 billion $Dynamic: +50% in the last 10 years

Recording Industry Association of America (RIAA)

Size: 12.3 billion $

Dynamic:

-2%

in the last 10 years (stagnant)

1. Introduction

10/25/2010

Slide4

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide5

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide6

Games’ computational model

Single sequential loop3 steps in each loop:Game-world state updateEntity interaction computation (dominant for

MMOGs

)

Entity state updates

Load generated by

(2)

is non-deterministic

 human factor

2. Model

GAME LOOP

Game-world update

Interaction computation

Entity states update

10/25/2010

Slide7

Game parallelization models

Zoning: huge game-world division into geographical sub-zones – each zone is handled by different machinesReplication: the same game-world handled by different machines, each one handling a subset of the contained entities (synchronized states)

Instancing:

multiple instances of the same zone with independent states. (World of

Warcraft

,

Runescape

,..)

10/25/20102. Model

Slide8

Dynamic resource provisioning

Main advantages:Significantly lower over-provisioningEfficient coverage of the world is possible10/25/20102. Model

Massive join

Massive join

Massive leave

Slide9

Static vs. Dynamic allocation

Reduce the resource waste for hosting MMOGs to as low as 10 times the one of static allocation10/25/20102. Model

250%

25%

What is the incentive for using dynamic allocation for operating

MMOGs

?

Slide10

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide11

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide12

Resource provisioning

Game operators: Predicted loads  requestsData centers and Cloud providers:

Time-space-(virtualization)

renting policy

offers

Resource allocation:

request – offer matching

Resource localityResource fitness

Resource bulk

size/proportionality

virtualization overhead

Time bulk size

10/25/2010

3. Method

OFFER

3

REQ

4

OFFER

12

VIRT

1

Slide13

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing10/25/2010

3. Method

Slide14

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide15

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide16

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide17

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide18

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide19

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide20

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide21

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide22

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide23

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide24

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide25

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide26

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide27

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide28

Load balancing

Possible actions:ReplicationClient migrationDe-replicationServer migrationInstancingDe-instancing

10/25/2010

3. Method

Slide29

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide30

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide31

Setup: Load balancing experiment

10/25/20104. ExperimentsDemonstrator applicationFirst Person Shooter – game type (FPS)Based on Real-Time Framework (RTF)Graphical interface utilises OGRESupports zoning and replication techniques

Testbed

7 machines from Amis (Slovenia) – game servers & resource allocation service

10 machines from University of Innsbruck –automated clients (bots)

Slide32

Demonstrator application

10/25/20104. Experiments

Slide33

Results: Load balancing experiment

10/25/20104. Experiments

Metric

Dynamic

res.

alloc

. method

Client

number threshold method

40 clients/server

50 clients/server

Under-allocation (avg.)

0.66%

0.86%

8.69%

Resource

utilisation

83.3%

100%

83.3%

Slide34

Setup: Cloud hosting experiment

Traces from the 2nd most popular MMOG, RuneScape1M paying accounts135M registered accounts since 20017M total playing (~6M free)Input:Trace composition:7 countries, 3 continentsMore than 130 game worldsConsisting of:Geographical locationNumber of clients

Over 10,000 samples at 2 min. interval, 2 weeks

Cloud resource types:

Amazon EC2 standard small $0.085/hour

Flexiscale

2GB $0.159/hour

NewServers Medium $0.170/hour

10/25/20104. Experiments

Report by Geoff

Iddison

Leipzig Games Convention

Slide35

Results: Cloud hosting experiment

LoadEstimated yearly MMOG hosting costs [$]

Amazon EC2

Flexiscale

NewServers

Dynamic

Static

Dynamic

Static

Dynamic

Static

0%

0

101,266

0

189,426

0

202,531

20%

23,326

101,266

40,920

189,426

38,468

202,531

50%

57,345

101,266

100,404

189,426

97,495

202,531

60%

57,830

101,266

101,829

189,426

98,179

202,531

70%

66,299

101,266

116,458

189,426

114,775

202,531

80%

75,709

101,266133,111189,426129,119202,53190%84,007101,266147,055189,426142,578202,53195%88,199101,266155,039189,426149,793202,531@ 50% average load: –47% yearly hosting expenses @ 90% average load: –23% yearly hosting expenses 10/25/20104. Experiments

Slide36

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide37

OUTLINE

IntroductionResource Provisioning ModelMethodExperiments

Conclusions

10/25/2010

Slide38

Conclusions

MMOG: application with >24 million user base Dynamic hosting modelInvestigated:Load balancing techniquesImpact of using Cloud resources on hosting expensesCurrent work: Investigating a new business model for MMOG operation10/25/2010

5. Conclusions

Slide39

edutain@grid

– http://edutaingrid.eu/on-going research – FWF project10/25/2010