Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.1

Part2: An Approach of Passive Peer based Caching to Mitigate P2P Inter-domain Traffic

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.2

Motivation

• Widespread of P2P file sharing applications– Traffic Increase: more serious in inter-domain links

• P2P Traffic– Undesirable characteristics compared with Web traffic

• Transmission data size is huge• No traffic centralized point, e.g. servers in a data center,

because communications will occur among peers at any locations

– Most of such huge data transmissions become inter-domain traffic

• P2P traffic volume is a critical issue for ISPs– Analyze P2P traffic impact on inter-domain traffic– Control P2P traffic on inter-domain links

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.3

Related Works

• P2P Traffic control– Filtering / limiting P2P traffic

• How do we identify P2P connections?– Deep packet inspection is very powerful but …– How deep is an ISP allowed to inspect the packets?

– Introducing asymmetry along physical network topology

• Network-aware clustering / Cache servers• Some application’s protocol is closed

More general control methods should be considered

• P2P Traffic model– Mainly focusing on the scalability of overlay network

• Number of hops and search messages on overlay networks We need a model with physical network viewpoint

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.4

Our Research

• Analysis of inter-domain traffic– Considering pure P2P file sharing applications– Modeling inter-domain traffic amount

• Influence of additional passive peer

• P2P traffic control using passive peer– Resource cache equivalent function on P2P network

with closed protocol

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.5

Traffic Flow Model : Terminology

• My Domain / Other Domains– Set(s) of peers in the same administrative scope (e.g.,

ISP)– My Domain : A domain of interest– Other Domains : Others

• Passive peer– A native peer for corresponding P2P applications

• without any change• no active operations (resource creation / resource request)

– Added to My domain

• We analyzed the influence of a passive peer addition

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.6

Traffic Flow Model : Traffic

• Traffic Flow Model– Modeling traffic incoming to / outgoing from passive peer– Four kinds of traffic

• ‘out’ / ‘in’ : outgoing / incoming

• ‘m’ / ‘o’ : My Domain / Other Domains

• Inter-domain Traffic : To out + To inInter-domain Traffic

Tmin

TmoutPassive Peer

Toout

Toin

Other DomainsMy Domain

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.7

Traffic Flow Model : Selection Probability (1/2)

• Selection Probability– Probability that a peer selects destination peers in My Domain– We assume :

p : Selection Probabilitiesq : 1 - p

Tm

inpqTo

in

Tm

out

To

out

Tmin

TmoutPassive Peer

Toout

Toin

Other DomainsMy Domain

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.8

Traffic Flow Model : Selection Probability (2/2)

• What does this equation mean ?– Destination selection is regarded as random selection on a domain level scale

Tm

inpqTo

in

Tm

out

To

out

p : Selection Probabilitiesq : 1 - p

Tmin

TmoutPassive Peer

Toout

Toin

Other DomainsMy Domain

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.9

Traffic Flow Model : Without Passive Peer

• If the passive peer were not added– : disappeared– : transferred to alternative peers

– Alternative peers are selected according to the selection probability p

Tmin

TmoutPassive Peer

Toout

Toin

Other DomainsMy Domain

ToinTmin ,Tmout , Toout

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.10

Traffic Flow Model : Traffic Amount Change

• Inter-domain Traffic :

• Ratio: r =

– Traffic amount change caused by the passive peer

pTo out + qT

m out

To out + To inpTo out + qTm out

pToout

pTmout

qTmout

qToout

Other DomainsMy Domain

Inter-domain Traffic

with passive peer

without passive peer

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.11

Traffic Flow Model : Analyze

• Area of decreasing inter-domain traffic by passive peer– If p < 1/2,– additional peer rather increases inter-domain traffic

20

40

60

80

100

0.0 0.2 0.4 0.6 0.8 1.0

CachePerformance

Selection Probabilities

r < 1(decrease)

r > 1(increase)

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.12

Proposal

• To decrease inter-domain traffic without change p,q– To decrease r less than 1.0

– r =

• Method : Filtering

To out + To inpTo out + qTm out

To

out

Other DomainsMy Domain

Tooutδ TooutFilter

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.13

Proposal Method : Analyze

• Effective areas for several filter ratio:δ– The smaller δ is, the less the inter-domain traffic becomes– The smaller δ enlarges the effective area

20

40

60

80

100

0.0 0.2 0.4 0.6 0.8 1.0

CachePerformance

Selection Probabilities

Filter : 0.5Filter : 0.2Filter : 0.1

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.14

Discussion

• Validity of traffic flow model– Is destination selection really random even on a

domain level scale?

• Effectiveness of proposed method– Can this method decreases inter-domain traffic ?

• Measurement of real P2P network traffic– In order to discuss the above two topics

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.15

Discussion : Simulation Analysis

• Selection Probability : p– Simulation with 200,000 peers’ behavior

• Selection Probability gradually converges– Common p value can be applied to both incoming and outgoing

communications

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.16

Discussion : Experimental Analysis

• Selection Probability : p– Calculated from incoming and outgoing traffic

• Selection Probability gradually converges

0.5

0.4

0.3

0.2

0.1

00 5 10 15 20

outgoingincoming

SelectionProbability

Time (days)

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.17

Discussion : Cache Effectiveness

• Ratio of inter-domain traffic change by passive peer– become less than 1 after 12th day

• Successfully decreases inter-domain traffic

22

20

6

4

2

00 5 10 15 20 25 30

Time (days)

Ratio ofTraffic Amount

Service Control Technologies for Peer-to-peer Traffic in Next Generation Networks p.18

Conclusion

• We discuss a novel traffic model of P2P file sharing Application– To grasp inter-domain P2P traffic trend– Passive peer concepts– Selection Probability

• Analyze an impact of P2P traffic on inter-domain traffic– Passive peer rather increases inter-domain traffic in

the case of p<1

• Propose P2P traffic control method– decrease inter-domain traffic even in the case of p<1