seminar 4: peer-to-peer systems - karlstad university · seminar on peer-to-peer systems m a r c e...

Sem

inar on Peer-to-Peer Systems

Mar

cel C

Cas

tro

Topics in Computer Networks 2010 • Seminar 4 1

Topics in Computer Networks 2010

Seminar 4:

Peer-to-Peer Systems

Marcel C. Castro

[email protected]

Sem


Mar

cel C

Cas

tro


Motivation

Internet Study

http://www.ipoque.com/resources/internet-

studies/

� Peer-to-Peer Systems:

� It is the dominating traffic type in the internet

� Based on application overlay routing

Protocol Type Distribution, Germany 2007

Sem


Mar

cel C

Cas

tro


Seminar 4: Peer-to-Peer Systems

�Client-Server concept

�Understanding P2P

�P2P and Wireless Networks

Sem


Mar

cel C

Cas

tro


Client-Server Model

� Simple Architecture

� Powerful central entity to host:� the whole resource index

� global management of resources (i.e. access rights etc)

� Low cost in messages for discovery

� Increase of clients may degrade the quality of service

Sem


Mar

cel C

Cas

tro


Why (not) Client-Server Model

�Advantages:

� Fast and guaranteed discovery (if resource exists)

� Easy to deploy and charge system-wide services

� Easy to retain resource consistency

� Facilitates configuration for maximum security of delivered services

�Weaknesses:

� Single point of failure

� High initial installation and maintenance cost

� Performance bottleneck – scalability issue

�Preferable in small environments

� Good in relatively predictable growth patterns.

Sem


Mar

cel C

Cas

tro


Seminar 4: Peer-to-Peer Systems




Sem


Mar

cel C

Cas

tro


Understanding P2P

� Issues addressed by P2P

� Eliminate the control that can be imposed on clients

� Provide high availability

� Provide scalability

� Provide privacy

� How to address them

� By distributing the logic and the resources

� By expanding horizontally not vertically (collaboration)

� By adapting to network changes on the fly

� By using direct client-to-client communication

Sem


Mar

cel C

Cas

tro


P2P definition

�A self-organizing, distributed

network of entities which

contribute their individual

resources and collaborate in order

to reach the goal for which the

network was built.

� P2P networks are those which exhibit 3 characteristics:

� self-organization

� distributed control / resources

� symmetric communication Dan Pascu, “Overview of P2P SIP

Principles & Technologies”. Int. SIP Conf.

2007.

Sem


Mar

cel C

Cas

tro


P2P Overlays

� No central entity: distribution among all nodes of� the whole resource index

� global resource management (i.e. access rights etc)

� Direct communication between nodes � Each node acts as both resource requestor and provider

� Application specific networks

� Increase of nodes may improve the quality of service

Sem


Mar

cel C

Cas

tro


Why (not) P2P Overlays

�Advantages

� Robust and fault-resilient architecture

� Low installation and maintenance cost

� Enables inexpensive resource redundancy

�Weaknesses

� Slower not always guaranteed discovery

� Frequent join/leave actions of nodes

� Heterogeneous node capabilities

� High discovery cost– scalability issue

� Distributed data storage is hard to do in a consistent way

�Suitable for large systems

� unpredictable growth patterns

Sem


Mar

cel C

Cas

tro


Overlay definition

� An overlay network is a virtual network of nodes and logical

links that is built on top of an existing network with the

purpose to implement a network service that is not available in

the existing network. (by I. Stoica)

� A P2P network is an overlay itself (over TCP/IP)

� E.g. of services:

• Routing (Resilient Overlay Networks, BGP routing, MPLS)

• Security (VPN)

• Application-level Multicast

Sem


Mar

cel C

Cas

tro


IP

Overlay networks

Overlay

Sem


Mar

cel C

Cas

tro


Overlay networks

IP

Overlay

Sem


Mar

cel C

Cas

tro


P2P Overlay Classification

� Level of Centralization

� Hybrid: node grouping with central index

� Pure: no centralised entity – equal nodes

� Resource Location

� Unstructured: any node may host any resource

� Structured: resources are hosted by well-defined nodes

G. Exarchakos , “P2P Overlays for

Scalable Service Access”. ICCGI’08.

Sem


Mar

cel C

Cas

tro


P2P overlay design

1. Choice of identifier space (IS)

- 128 bits overlay identifier

2. Map resources and peers to IS

- HASH(MAC, IP, or Locality)

3. Management of the IS by

peers (or Resource Location)

- Unstructured, structured

4. Routing strategy

- Flooding, random walk, Finger table

(Chord)

5. Maintenance strategy

- Proactive (probe), or reactive

correction

Sem


Mar

cel C

Cas

tro


Management of the Identifier Space

�Local (Unstructured P2P)� Each peer only indexes its own content and flood queries widely

� Can perform complex searches (rich queries not just key lookups

� Blind techniques: random forwarding from node to node.

e.g.: Gnutela

Sem


Mar

cel C

Cas

tro



�Unstructured P2P

�Advantages

� Fast lookup

� Low join and leave overhead

� Popular files are replicated many times

�Weaknesses

� Not 100% success rate

� Very high communication overhead

� Uneven load distribution

Sem


Mar

cel C

Cas

tro



�Distributed (Structured P2P)

� Also known as Distributed Hash Table (DHT)

� Efficient key lookup / routing (no flooding)

� Can perform only exact key lookups

� Many recent academic systems –

• CAN, Chord , Kademlia, Pastry, Tapestry, Viceroy, Bamboo.

Sem


Mar

cel C

Cas

tro


IDEA:

�Route a packet based on a key to the node in the

network that is currently responsible for the packet's

key.

� This process is referred to as indirect or key-based

routing.

DHT - Distributed Hash Table

Sem


Mar

cel C

Cas

tro


K6

K46

K36

N1

N8

N14

N21

N32

N38

N42

N48

N51

� Large id space

DHT - Distributed Hash Table

root nodefor key

id space

nodeId

key

64 |0 (6 bits)

�NodeIds picked randomly from space

�Hash[192.168.1.1] = N1

�Keys picked randomly from space

�Hash[picture.jpg] = K46

�Key is managed by its root node:

� Live node with id closest to the key

� location of object or

actual object

Overlay

IPIP

Sem


Mar

cel C

Cas

tro


Key-based routing

K46

N1

N8

N14

N21

N32

N38

N42

N48

N51

nodeId

key

Who’s respfor file K46

file K46stored here

I am

O(N) messages

on avg to resolve

query

Sem


Mar

cel C

Cas

tro


The Chord DHT overlay

� Uses SHA-1 hashes (160 bits)

� Maps nodes and keys to a ring

� O(log N) lookup performance

� O(log N) routing table size

� Supports join and leave operations

for maintaining the network

� It basically supports one operation:

lookup a node for a given key

� Each node handles the resources which have their hashes mapped between the

node itself and its predecessor

file K46 ?

� Each node knows its predecessor, successor and keeps a list of successor nodes

known as the finger table, which is used to improve lookup performance and increase

fault tolerance

� If a lookup doesn’t yield a local resource, it is forwarded to the node in the finger

table which has the closest hash value preceding the hash of the queried resource

I. Stoica, et. al., “Chord: A Scalable Peer-topeer

Lookup Service for Internet Applications”, SIGCOMM’01

Sem


Mar

cel C

Cas

tro


Tradeoff of routing table size

vs. network diameter

Full Mesh

Ring

Chord

Sem


Mar

cel C

Cas

tro



�Structured P2P

�Advantages

� Scalability: O(log N) routing

� Load-balancing

� Overlay robustness

�Weaknesses

� No control where data is stored

� Complex queries are not possible

� Join and leave overhead

Sem


Mar

cel C

Cas

tro


Outline




Sem


Mar

cel C

Cas

tro


Challenges of P2P in Wireless Environments

�Differences between the Internet and a Mobile Wireless Networks:

� Bandwidth limitation

� Multi-access interference

� Node mobility

• Churn (frequent node joining and leaving the network)

� Limited energy

� State-efficiency trade-off

Sem


Mar

cel C

Cas

tro


Example: P2P and Mobile Ad-hoc Network

Physical, to all nodes in

transmission range area

Virtual, multiple unicastsBroadcast

MobileFixedPeer Mobility

Proactive, reactive,

hybrid


hybrid

Routing

Physical structure

corresponding to logical

structure

Physical apart from

logical structure

Structure

Restricted areaAny Internet pointPeer location

Low ( wireless

connections)

High ( physical

connections)

Connection

confidence

Wireless and indirectFixed medium and directConnection Between

two nodes

A physical infrastructure

to provide connectivity

Logical infrastructure to

Provide a service

Motivation for

Creating the network

MANETP2P Networks

Physical, to all nodes in

transmission range area

Virtual, multiple unicastsBroadcast

MobileFixedPeer Mobility


hybrid


hybrid

Routing

Physical structure

corresponding to logical

structure

Physical apart from

logical structure

Structure

Restricted areaAny Internet pointPeer location

Low ( wireless

connections)

High ( physical

connections)

Connection

confidence

Wireless and indirectFixed medium and directConnection Between

two nodes

A physical infrastructure

to provide connectivity

Logical infrastructure to

Provide a service

Motivation for

Creating the network

MANETP2P Networks

Sem


Mar

cel C

Cas

tro


How to integrate P2P and MANET

Key-based

routing MANET

(DHT routing)

Broadcast,

on demand

routing (AODV)

MANET

Routing

Structured,

DHT

Unstructured,

FloodingP2P techniques

1

Sem


Mar

cel C

Cas

tro


ORION: Unstructure P2P over MANET

Qu

ery

Ph

ase

Re

spo

nse

P

ha

se

A. Klemm et al, “A special-purpose peer-to-peer file

sharing system for mobile ad hoc networks”.VTC 2003

1. Orion:

� Broadcast-based P2P lookup protocol over MANET on-demand routing protocols.

• Easy to implement

• Scalability problem

Strict layering of unstructured overlay approached on

top of wireless routing protocols is unlikely to work

in MANETs.

Sem


Mar

cel C

Cas

tro


How to integrate P2P and MANET

Key-based

routing MANET

(DHT routing)

Broadcast,

on demand

routing (AODV)

MANET

Routing

Structured,

DHT

Unstructured,

FloodingP2P techniques

2

Sem


Mar

cel C

Cas

tro


Bamboo/AODV: Structured P2P over MESH

2. Bamboo/AODV:� DHT-based P2P protocol

over MANET.

� P2P lookup can be scalable

� Underlying network routing protocol is based on broadcast

� Complexity of routing algorithm

• O(nlog n)

Fig. 1: : Impact of Bamboo Management traffic

0

200

400

600

800

1000

1200

4 9 16 25 36 49

Number of Nodes

To

tal O

verl

ay T

raff

ic (

Kb

ytes

)

NO management

CUSTOM management

STANDARD management

0,00

10,00

20,00

30,00

40,00

50,00

60,00

70,00

80,00

90,00

100,00

4 9 16 25 36 49

Number of Nodes

Su

cces

s R

atio

(%

)

NO management CUSTOM management STANDARD management

Fig. 2: Request Success rate

IPIP

A

B

A

B

OverlayOverlayA BA B

- Flooding is still necessary to discover routes and

maintain the DHT.

M. Castro et al, “Performance Evaluation of Structured

P2P over Wireless Multi-hop Networks”. MESH 2008

Sem


Mar

cel C

Cas

tro


Oracle: Example of Overlay/Underlay interactions

� Oracle provided by an ISP

� Information is provided, like link delay, bandwidth estimations, etc.

� As input to the Oracle, a list of P2P nodes sharing a known content is given

and the list ranked according to different performance metrics is returned

from Oracle

� Results

� P2P applications won’t have to

perform such measurements by

themselves

� The Oracle gives ISPs a way to

control overlay routing

� Underlay to overlay

information exchange

improves overlay operations

and underlay network usage

Sem


Mar

cel C

Cas

tro


Oracle: Example of Overlay/Underlay interactions

� Oracle provided by an ISP

� Information is provided, like link delay, bandwidth estimations, etc.

� As input to the Oracle, a list of P2P nodes sharing a known content is given

and the list ranked according to different performance metrics is returned

from Oracle

� Results

� P2P applications won’t have to

perform such measurements by

themselves

� The Oracle gives ISPs a way to

control overlay routing

� Underlay to overlay

information exchange

improves overlay operations

and underlay network usage

ALTO IETF Working Group:

(Application-Layer Traffic

Optimization )

Standardize information exchange

between applications and ISPs in order

to help in peer selection

ALTO IETF Working Group:

(Application-Layer Traffic

Optimization )

Standardize information exchange

between applications and ISPs in order

to help in peer selection

Sem


Mar

cel C

Cas

tro


Caching: Example of Overlay/Underlay interactions

� Caching key-lookups to reduce

routing stretch:

� How much information can be

used by a node to reduce the

routing stretch?

• Use of current 2-hop neighbor

information in the hello

messages together with cache

information;

• How to gather Cache

information ?

– Key’s request that passes

through a node;

– Key’s request that nodes can

hear in its transmission range

(cross-layer information).

1000 Nodes: Stretch versus Node Density

1

1,5

2

2,5

3

3,5

4

~10 ~13 ~22 ~40

Node Density

Str

etch

default

NoN

NoN+Cache256

NoN+Cache-1

M. Castro et al, “Minimizing DHT Routing Stretch in

MANETs”. Adhoc'09, Sweden.

Sem


Mar

cel C

Cas

tro


Georoy: Example of Overlay/Underlay interactions

� Georoy: a location-aware DHT for WMNs

� Mapping P2P Overlay to physical topology

• Able to obtain a small stretch factor

� It combines an unit ring topology and a butterfly

network topology

� Lookup performance of O(log n)

� Routing Table containing at most 7 entries

� Two-tier architecture

� Super Peers (SPs): provide

distributed resource

catalog.

� Leaf Peers (LPs): share and

request resources by

querying their associated

super peers.

� Home SP: responsible for

managing the point to the

LP’s physical location.

Leaf Peer

Super Peer

s - square region side R=[0,s)2

(x,y) - coordinate of peer in R

∆ – constant with 0 < ∆ < s

Sem


Mar

cel C

Cas

tro


Georoy: Example of Overlay/Underlay interactions

� Simulations: ns-2

� Grid topology

� TwoRayGround propagation model

� AODV-UU routing

� Avg. of 100 random key lookups

� Comparison:

� Bamboo: better numb. of logical hops by

exploiting leafset information and routing

tables -> fast ring traversal

� Georoy: close-fitting mapping between

logical and physical hops

� Bamboo: small lookup latency for smaller

network size due to routing table

information

� Georoy: small lookup latency as number of

nodes increase -> logical and physical

topologies tightly coupled

• logical path is not very different from the

physical path.

M. Castro et al, “On the comparison between

performance of DHT-based protocols for opportunistic

networks”. FUNEMS, Italy, 2010.

Sem


Mar

cel C

Cas

tro


Overlay networks in Wireless Context

� Interaction between Overlay networks and Underlay required!

� To enable ISPs to perform efficient traffic engineering by enabling traffic

control

� To enable overlays to create better overlay structure/topology

� To reduce overhead as overlay probe network for bandwidth

availability/delay

� Example: Wireless Mesh Networks as Underlay

� Increases the problem due to interference and control traffic overhead

� Multi-Channel operation required to allow higher capacity

� Internet Gateways are a strong topology element � could optimize the

overlay

� Need to engineer channel assignment to match with traffic demands and

topology

� Cross-layer approach can create an information exchange between

Overlay and Mesh Underlay

Sem


Mar

cel C

Cas

tro


References

1. M. Gerla and C. Lindemann and A. Rowstron. “P2P MANET's - New Research Issues”. Perspectives Workshop: Peer-to-Peer Mobile Ad Hoc Networks -New Research Issues, 2005.

2. Thomas C. Zahn, “Structured Peer-to-Peer Services for Mobile Ad Hoc Networks”, PhD Thesis FU Berlin, 2006.

3. Jorg Eberspacher et al., “Structured P2P Networks in Mobile and Fixed Environments”HET-NETs04.

4. I. Stoica, R. Morris, D. Liben-Nowell, D. R. Karger, M. F. Kaashoek, F. Dabek, and H. Balakrishnan. “Chord: A scalable peer-to-peer lookup protocol for internet applications”. IEEE/ACM Transaction Network, 11(1):17–32, 2003.

5. M. Caesar, M. Castro, E. B. Nightingale, G. O’Shea, and A. Rowstron. “Virtual Ring Routing: Network routing inspired by DHTs”. In ACM SIGCOMM’06, pages 351-362, 2006.

6. T. Fuhrmann, P. Di, K. Kutzner, and C. Cramer , “Pushing Chord into the Underlay: Scalable Routing for Hybrid MANETs”, Technical Report, Universität Karlsruhe (TH) 2006-12, June 2006.

Sem


Mar

cel C

Cas

tro


References

7. E. K. Lua, J. Crowcroft, M. Pias, R. Sharma, and S. Lim. “A survey and comparison of peer-to-peer overlay network schemes”. IEEE Communications Surveys, 7(2):72–93, 2005.

8. A. Klemm et al, “A special-purpose peer-to-peer file sharing system for mobile ad hoc networks”.VTC 2003.

9. K. Aberer, et al. The essence of P2P: a reference architecture for overlay networks. Fifth IEEE International Conference on Peer-to-Peer Computing, Sep 2005, Konstanz, Germany.

10. Y. Charlie Hu, Saumitra M. Das, and H. Pucha, "Peer-to-Peer Overlay Abstractions in MANETs“. Book Chapter in Theoretical and Algorithmic Aspects of Sensor, Ad Hoc Wireless, and Peer-to-Peer Networks,CRC Press, 2005.

11. T. Fuhrmann, “Performance of scalable Source Routing in Hybrid MANETs”. In Wireless on Demand Network Systems and Services (WONS), 2007.

12. L. Galluccio, G. Morabito, S. Palazzo, M. Pellegrini, M. E. Renda, and P. Santi. Georoy: A location-aware enhancement to viceroy peer-to-peer algorithm. Computer Networks, 2007.

Sem


Mar

cel C

Cas

tro


Thank you!

Marcel C. Castro

[email protected]

www.cs.kau.se/~marccava

Sem


Mar

cel C

Cas

tro


MANET - Mobile Ad-hoc Networks

�Mobile Ad-hoc Networks (MANET) consist of mobile

nodes communicating with each other using multi-hop

wireless links.

�Self-organization for communication.

�Self-healing to cope with network failures.

�Without (necessarily) using

a pre-existing infrastructure.

Sem


Mar

cel C

Cas

tro


Mandatory Reading

�"A survey and comparison of peer-to-peer overlay

network schemes” E. Keong Lua, J. Crowcroft, M. Pias,

R. Sharma, and S. Lim, IEEE Communications Surveys &

Tutorials, Second Quarter 2005, Volume: 7, Issue: 2 pp

72-93.

�"Theoretical and Algorithmic Aspects of Sensor, Ad Hoc

Wireless, and Peer-to-Peer Networks”. Y. C. Hu, S. M.

Das, and H. Pucha, CRC Press, 2005, ch. Peer-to-Peer

Overlay Abstractions in MANETs, pp. 858–871.

seminar 4: peer-to-peer systems - karlstad university · seminar on peer-to-peer systems m a r c e...

Documents