UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE STRUCTURED P2P OVERLAY NETWORKS Yann Busnel1Structured
P2P overlay networks Yann Busnel Maitre de confrence Universit de
Nantes F2R401b - Architectures des systmes dynamiques Telecom
Bretagne Rennes Slide 2 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Recall on unstructured P2P
Unstructured P2P architectures Direct application of P2P paradigm
Each peer: Is both client and server Is autonomous in term of
storage and resources Has only a knowledge of its logical
neighborhood Problems : Numerous limitations due to The number of
exchanges messages The efficiency of localization process Blind
approach Yann Busnel2Structured P2P overlay networks Slide 3
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Polemic P2P paradigm is realistic ? Nodes from a network are they
all really equivalent? Is it reasonnable to not indexing resources
in order to protect autonomy? Yann Busnel3Structured P2P overlay
networks Slide 4 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Constat 1 All peers are not equivalents:
Bandwidth heterogeneity A gap from 1 to 3 of the available
bandwidth [Sar02] Storage capabilities heterogeneity Processor
heterogeneity Behavior heterogeneity In file sharing system, 70% of
users are free riders i.e. users that dont share any file [Ada00]
50% of replies are produced from 1% of peers [Ada00] Finally, all
peers are not equals! Motivation for HYBRIDES P2P ARCHITECTURES
Yann Busnel4Structured P2P overlay networks Slide 5 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Constat 2
Resources indexation remains (until now) the best way to guarantee
resource localization Finally, in order to maintain a good
localization service QoS, it is not reasonnable to avoid resource
indexes! Motivation for STRUCTURED P2P ARCHITECTURES Yann
Busnel5Structured P2P overlay networks Slide 6 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Outline Introduction
Hybrid P2P Structured P2P DHT Interface Example system CHORD PASTRY
CAN Yann Busnel6Structured P2P overlay networks Slide 7 UNIVERSIT
DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Hierarchical peers Remove the peers equality constraint Aims: Give
to some peers a specific role Representation role Routing role
Remove low bandwidth nodes from the localization process: ask for
other peers to represent them Strategies Creation of peers data
index on a subset of peers => Super-peer Yann Busnel7Structured
P2P overlay networks Slide 8 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Super-peer Super-peer / Peer
Super-Peer : Nodes with a large bandwidth Peer : Node with a low
bandwidth are clustered around a super-peer Role: To substitute
itself to simple peers in the localization process Indexation
Super-peers have an index of resources provided by represented
peers. Note that Napster : 1 super-peer / n pairs KaZaA : m
super-peers / n pairs Yann Busnel8Structured P2P overlay networks
Slide 9 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Hierarchical network Super-Peers are organized in P2P
fashion All Peers are linked to a super-peer in client/serveur mode
pair super-pair C / S P2P i i i i i Index des ressources du cluster
cluster Yann Busnel9Structured P2P overlay networks Slide 10
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Download Peer architecture C/S Communication Peer User Interface
Request Locales resources Result Request Download service Resource
Super-peer connection system Super-peer interogation system Index
of stored data Index Super-Peer @ super peer Yann
Busnel10Structured P2P overlay networks Slide 11 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Download
Super peer architecture Message management Super-Peer User
Interface Request Resources locales Super-Peer Neighborhood table
Request management Request handler Result transmitter Request
forwarder Result handler Results Result Request Request Request
Request Downloading service Resource Index of local data and
represented peers data C/S Comm. C/S communication server Results
Peer Yann Busnel11Structured P2P overlay networks Slide 12
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Fault tolerance Redondant Super-peer Possible enhancement, choose k
super-peers (partners) instead of one in a cluster Each partner is
connected to each client and owns an index for all their resources
Raise join cost of new node by a factor k Clients send their
requests to their partner according to a round robin principle
Permit to decrease the load of a partner by a factor k Yann
Busnel12Structured P2P overlay networks Slide 13 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Example of
redondant super-peer C/S P2P Yann Busnel13Structured P2P overlay
networks Slide 14 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Remarks R1 : increase the cluster size raise
individual load (in term of bandwidth) of super-peers but reduce
the global load Large clusters => Huge consequence in term of
failure, attacks, R2 : super-peer redondance is positive but
costful Good trade-off between global load balancing using large
cluster with low individual load and a good ault tolerance Yann
Busnel14Structured P2P overlay networks Slide 15 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Summary
Avantages Remove low resource peers (bandwidth, processing power, )
from the information localization process Faster resource
localization Disadvantages Creation of an index on super-peer
likely to resolve queries Centralization, even local, infers a
lower super-peer failure tolerance Costly super peer redondance
Yann Busnel15Structured P2P overlay networks Slide 16 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Example of
hierarchical P2P Super-peer : KaZaA 4 500 000 simultaneous
connections Connexions duration: few hours Number of available
files: 10 6 Gnutella (0.6-*) Ultrapeers Yann Busnel16Structured P2P
overlay networks Slide 17 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Outline Introduction Hybrid P2P
Structured P2P DHT Interface Example system CHORD PASTRY CAN Yann
Busnel17Structured P2P overlay networks Slide 18 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Objectives
Give value to efficiency of localization process Data indexation
Guarantee to find any available data in the network Required a
decentralized index ! Yann Busnel18Structured P2P overlay networks
Slide 19 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Structure networks Properties Data or data reference are
placed on specific nodes Allocation are made using a hash function
Searching is proceed according to a key Challenge Find fastly an
object starting from its key in a large scale network without
centralized structure Chord, CAN, Pastry Yann Busnel19Structured
P2P overlay networks Slide 20 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE 1-P2P routing infrastructure
Overlay: network abstraction on top of IP Basic functionality:
distributed hash table CAN relies on a multidimensional cartesian
space Chord, Pastry, Tapestry: generalized hypercube routing based
on prefix matching Exact-match interface Applications
Content-delivery networks Storage systems, Caching Naming services
Multicast Yann Busnel20Structured P2P overlay networks Slide 21
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Distributed Hash Table (DHT) Yann Busnel21 k6,v6 k1,v1 k5,v5 k2,v2
k4,v4 k3,v3 containers Operations: insert(k,v) lookup(k) Table of
containers Store pairs Efficient access to a value given a key
Mapping key-value ensured by the table of containers Structured P2P
overlay networks Slide 22 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE P2P Distributed Hash Table Yann
Busnel22 k6,v6 k1,v1 k5,v5 k2,v2 k4,v4 k3,v3 nodes Operations:
send(m,k) P2P overlay network Message sent to keys: implementation
of a DHT P2P infrast. ensures mapping between keys and physical
nodes Fully decentralized: peer to peer communication paradigm
Structured P2P overlay networks Slide 23 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE DHT capabilities
Scalability: Resources agregation Storage Computing power Bandwidth
Load balancing Availability Self-organizing Failure resilience Yann
Busnel23Structured P2P overlay networks Slide 24 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE 24 P2P DHT
principle 52 24 75 40 18 91 32 66 83 Yann Busnel24Structured P2P
overlay networks Slide 25 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE logical address space 52 24 75
40 18 91 32 66 83 South America North America Australia Asia Europe
Asia Overlay network P2P DHT principle Yann BusnelStructured P2P
overlay networks25 Slide 26 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Mapping Yann Busnel26 Identifier
Space key nodeId An identifier is associated to each node (nodeId)
Large identifier space (keys and nodeId) A node is responsible for
closest key to its nodeId Structured P2P overlay networks Slide 27
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
key nodeId Sending messages to keys Yann Busnel27 source
destination Each node maintains a routing table (nodeid, IP adress)
At each routing hops, the message gets closest to the key
Infrastructure P2P: mapping between key and node Structured P2P
overlay networks Slide 28 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Distributed Hash Table Yann
Busnel28 K V (K 1,V 1 ) retrieve (K 1 ) insert (K 1,V 1 )
Structured P2P overlay networks Slide 29 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Routing Interface
Route (key,node,msg) Send (node,msg) NextHopSet = lookup (key, num)
Upcalls Deliver(key,msg) Forward (key,msg, NexHopNode) Update
(neighbour set) Yann Busnel29Structured P2P overlay networks Slide
30 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Routing Interface Route (key, node, msg) Route the msg
to the node (alive) responsible of the key. Non blocking, (best
effort), the message may be lost or duplicated node parameter is
optional and is interpreted as the first hop (using a good
heuristic might deliver the message in one single hop) msg:
application-specific key: numerical key (160 bits) node:
encapsulates the @ of the last node known associated to that key
Yann Busnel30Structured P2P overlay networks Slide 31 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Routing
Interface nextHopSet = lookup (key,num)Returns a set of nodes (num)
from the routing table as potential next steps for the routing key:
numerical 160 bit key num: number of nodes required in the set
Result: the resulting set of nodes Yann Busnel31Structured P2P
overlay networks Slide 32 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Routing Interface Send(node,msg)
sends a message to a specific node; fails if the node is not
available, blocking and reliable Msg: msg is application-specific
Node: encapsulate the @ of the destination node Result: true if the
msg is delivered false otherwise Yann Busnel32Structured P2P
overlay networks Slide 33 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Routing Interface Deliver
(msg,key) Upcall delivering the message to application. Assumes one
application per node msg: message application-specific key:
numerical 160 bit key Yann Busnel33Structured P2P overlay networks
Slide 34 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Routing Interface Forward (msg,key,nextHopNode) upcall
invoked at each node on the message route (source and destination
included) to inform an application that the msg and the key are
about to be forwarded to NextHopNode. The application can modify
msg, key, nextHopNode or terminate the message by setting
nextHopNode to null Msg: application specific msg Key: numerical
160 bit key nextHopNode: @ of the node to which the message is
forwarded after the upcall has returned Yann Busnel34Structured P2P
overlay networks Slide 35 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Routing Interface
Update(neighborSet) upcall invoked to inform the application of a
change in the neighborhood of the local node. Neigborset: set of k
nodes Yann Busnel35Structured P2P overlay networks Slide 36
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Outline Introduction Hybrid P2P Structured P2P DHT Interface
Example system CHORD PASTRY CAN Yann Busnel36Structured P2P overlay
networks Slide 37 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Pastry / Chord Yann Busnel37 node key Id space
NodeId = 128 bits Nodes and key place in a linear space (ring)
Mapping : a key is associated to the node with the numerically
closest nodeId to the key (different placement in Pastry and Chord)
Structured P2P overlay networks Slide 38 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Chord DHT-based on a
ring that permit place 2 m nodes Node allocation Hashing of node IP
Each node is allocated on the ring according to its IP hash m = 4
16 emplacements m = 4 16 emplacements Yann Busnel38Structured P2P
overlay networks Slide 39 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Resource hash H(resource) = key
The key is placed on the node with ID greater or equal than key
value Resource allocation ( cl=0000, contenu=F1 ) ( cl=1001,
contenu=F3 ) ( cl=0010, contenu=F2 ) ( cl=0110, contenu=F4 ) Yann
Busnel39Structured P2P overlay networks Slide 40 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Naive
resource searching Each node store ID of its successor To handle a
query, a peer forward the request to its successor The result
follow the backward path Linear searching according to the number
of peers Yann Busnel40Structured P2P overlay networks Slide 41
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Example of naive searching Recherche(0000) ( cl=0000, contenu=F1 )
Yann Busnel41Structured P2P overlay networks Slide 42 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Research
enhancement Required a routing table more complete: Finger nodes
table Use a successor mechanism not limited to the direct one on
the ring Principle For each node Ni with hash value j :
Succ[k]=first node on the ring that verify (j + 2k-1) mod 2 m, 1 k
m Use of predecessor for ring maintenance Yann Busnel42Structured
P2P overlay networks Slide 43 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Yann BusnelStructured P2P
overlay networks Finger nodes table 1+2 0 =2N8 1+2 1 =3N8 1+2 2
=5N3 1+2 3 =9N5 10+2 0 =11N4 10+2 1 =12N4 10+2 2 =14N7 10+2 3 =2N8
[11,12 [ [12,14 [ [14,2 [ [2,10 [ [2,3 [ [3,5 [ [5,9 [ [9,1 [ 43
Slide 44 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Searching algorithm If the key exists locally Yes return
associated value Else Search in the routing table a node with a
value lower or equal than the searched key Forward the query to the
selected node Apply this recursively Average number of hops: O(log
2 (N)) Yann Busnel44Structured P2P overlay networks Slide 45
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
45Structured P2P overlay networks Searching in Chord Look for key
00001 from peer N2 Stored keys {00111,10111} Routing table (Finger
nodes) Stored keys {00111,10111} Routing table (Finger nodes)
ValueResp.Node 7 + 2 0 = 8[8,9[N5 7 + 2 1 = 9[9,11[N5 7 + 2 2 =
11[11,15[N4 7 + 2 3 = 15[15,7[N1 0 6 5 15 4 8 10 N5 9 7 1 3 2 12 11
13 14 N6 N8 N3 N2 N4 N7 N10 Stored keys {00000,01111,10000,11111}
Routing table (Finger nodes) Stored keys {00000,01111,10000,11111}
Routing table (Finger nodes) 00001 ValueResp.Node 0 + 2 0 =
1[1,2[N6 0 + 2 1 = 2[2,4[N8 0 + 2 2 = 4[4,8[N3 0 + 2 3 = 8[8,0[N5
Yann Busnel Slide 46 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE
DE NANTES ATLANTIQUE Node insertion Combining elementary
operations: N.JOIN(N) Node N tells to node N its arriving int the
network and ask for a successor N.STABILIZE() Periodically launch,
permit to N and its successor to check if they form a correct
couple (no new node inserted between them) N.MAJENTRY() :
Periodically launch, permit to initialize routing table of new
nodes, or update existing ones N.TESTPREC() : Periodically launch,
check if predecessor remains connected Yann Busnel46Structured P2P
overlay networks Slide 47 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Insertion of node N11 at
position 9 Insertion of node N11 at position 9 N11 1) Contact
successor 01000, 11000, 11001 9 >9 9 9 >9 2) Key retrival
(join) 3) Predecessor retrival (stabilize) 4) Routing table update
01010, 110010 Yann Busnel47Structured P2P overlay networks Slide 48
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
Insertion of node N11 at position 9 Insertion of node N11 at
position 9 9 9 N11 1) Contact successor 2) Key retrival (join) 3)
Predecessor retrival (stabilize) 4) Routing table update 01010,
110010 N5, N4, N10, N6 N11, N11 N11 N11 insertion generates
modification on 5 nodes 01000, 11000, 11001 Yann Busnel48Structured
P2P overlay networks Slide 49 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Insertion algorithm properties
Guarantee that any sequence of join interlace with stabilize
converges to a stable state (all nodes are reachable) Even in
presence of (limited) joins, searching cost remains in O(log(N)) A
search can fail if the network is not completelly stabilized (its
possible to send again the query a little bit later) Yann
Busnel49Structured P2P overlay networks Slide 50 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Failure
tolerance The algorithm relies of successor notion (Failure on one
node cut the network) A solution consists to manage a set of r
successor (cf. Pastry) Yann Busnel50Structured P2P overlay networks
Slide 51 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Chord evaluation Searching type: equality Searching
cost: O(log(n)) Update cost: O(log(n)) Node insertion cost: O(log 2
(n)) No storage or routing autonomy Yann Busnel51Structured P2P
overlay networks Slide 52 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Chord summary Simple algorithm,
with good demonstrable properties Confirmation by experimentation
Latency problem Searching function minimize hops but all hops dont
have the same physical cost (cf. Pastry) Required to use
information on distance between nodes (choose the closest successor
among the set of possible ones) Yann Busnel52Structured P2P overlay
networks Slide 53 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Pastry Slide 54 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Pastry (MSR/Rice) Naming space :
Ring of 128 bit integers nodeIds chosen at random Key/node mapping
key associated to the node with the numerically closest node id
Routing table: Identifiers are a set of digits in base 16 Matrix of
128/4 lines and 16 columns routeTable(i,j): nodeId matching the
current node identifier up to level I with the next digit is j Leaf
set 8 or 16 closest numerical neighbours in the naming space Yann
Busnel54Structured P2P overlay networks Slide 55 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Pastry:
Routing table(#65a1fcx) Yann Busnel55 log 16 (N) lines log 16 (N)
lines Line 0 Line 1 Line 2 Line 3 Structured P2P overlay networks
Slide 56 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Pastry: Routing Yann Busnel56 Properties log 16 N hops
Size of the state (routing table): O(log N) d46a1c Route(d46a1c)
d462ba d4213f d13da3 65a1fc d467c4 d471f1 Structured P2P overlay
networks Slide 57 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Routing algorithm, notations Yann
Busnel57Structured P2P overlay networks Slide 58 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Routing
algorithm (on node A) Yann Busnel58Structured P2P overlay networks
Slide 59 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Node departure Explicit departure or failure Replacement
of a node The leafset of the closest node in the leafset contains
the closest new node, not yet in the leafset Update from the
leafset information Update the application Yann Busnel59Structured
P2P overlay networks Slide 60 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Failure detection Detected when
immediate neighbours in the name space (leafset) can no longer
communicate Detected when a contact fails during the routing
Routing uses an alternative route Yann Busnel60Structured P2P
overlay networks Slide 61 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Fixing the routing table of A
Repair Yann Busnel61Structured P2P overlay networks Slide 62
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
State maintenance Leaf set is aggressively monitored and fixed
Routing table are lazily repaired When a hole is detected during
the routing Periodic gossip-based maintenance Yann
Busnel62Structured P2P overlay networks Slide 63 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Reducing
latency Random assignment of nodeId: Nodes numerically close are
geographically (topologically) distant Objective: fill the routing
table with nodes so that routing hops are as short (latency wise)
as possible Topological Metric: latency Yann Busnel63 d467c4 d467f5
6fdacd Structured P2P overlay networks Slide 64 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Exploiting locality
in Pastry Neighbour selected based of a network proximity metric:
Closest topological node Satisfying the constraints of the routing
table routeTable(i,j): nodeId corresponding to the current nodeId
courant up to level i next digit = j nodes are close at the top
level of the routing table random nodes at the bottom levels of the
routing tables Yann Busnel64Structured P2P overlay networks Slide
65 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Proximity routing in Pastry Yann Busnel65 d46a1c
Route(d46a1c) d462ba d4213f d13da3 65a1fc d467c4 d471f1 Naming
space d467c4 65a1fc d13da3 d4213f d462ba Topological space Leaf set
Structured P2P overlay networks Slide 66 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Locality Node X
routes to node A Path A,B, -> Z Z numerically closest to X
Initialisation of the line i of the routing table with the contents
of line i of the routing table of the ith node encountered on the
path Improving the quality of the routing table X asks to each node
of its routing table its own routing state and compare distances
Gossip-based update for each line (20mn) Periodically, an entry is
chosen at random in the routing table Corresponding line of this
entry sent Evaluation of potential candidates Replacement of better
candidates New nodes gradually integrated Yann Busnel66Structured
P2P overlay networks Slide 67 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Node insertion in Pastry Yann
Busnel67 d467c4 65a1fc d13da3 d4213f d462ba Topological space New
node : d46a1c d46a1c Route(d46a1c) d462ba d4213f d13da3 65a1fc
d467c4 d471f1 Naming space Structured P2P overlay networks Slide 68
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
68 04F2 5230 834B C52A 8909 8BB2 3A79 8954 8957 AC78 895D E25A 04F2
3A79 5230 834B 8909 8954 8957 8BB2 AC78 C52A E25A k = 8958 k = 8959
put(8959,block) root of key 8959 block replica 895D replica
Resource insertion in Pastry Yann Busnel68Structured P2P overlay
networks Slide 69 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE 69 04F2 5230 834B C52A 8909 8BB2 3A79 8954 8957
AC78 895D E25A 04F2 3A79 5230 834B 8909 8954 8957 8BB2 AC78 C52A
E25A block replica 895D replica k = 8958 k = 8959 get(8959,block)
Searching in Pastry in detail Yann Busnel69Structured P2P overlay
networks Slide 70 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Node discovery: approximation of the closest node
discover(seed) //destination nodes=getleafset(seed) //probes on the
leaf set nearNode=pickClosest(nodes) //pick closest in the leafset
depth=getMaxRoutingTableLevel(nearNode) //highest line in RT
closest=nil while (closest != nearNode) closest=nearNode nodes =
getRoutingTable(nearNode, depth) nearNode =pickclosest(nodes) if
(depth >0) depth = depth 1 end return closest Routing table
property used to move exponentially close to the closest Pick the
closest node at each level and get the the next level from it
(bottom up) Constant number of probes at each level Probed nodes
get exponentially closer at each level Fill routing table entries
from the closest node Yann Busnel70Structured P2P overlay networks
Slide 71 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Performance Yann Busnel71 1.59 slower than IP on average
Structured P2P overlay networks Slide 72 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Content Adressable
Network (CAN) Slide 73 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Content Adressable Network -CAN
Virtual coordinates Cartesian space The space is shared between
peers Each node is responsible for a part of the space (zone)
Abstraction CAN enables to store data at a given point in the space
CAN enable the routing from a point of the space to the other (DHT
functionality) A point is associated to the node owning the zone in
which the point lies Yann Busnel73Structured P2P overlay networks
Slide 74 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Space organisation in CAN Yann Busnel74 node Key Name
space D-dimension space Routing: progression within the space
towards The destination Structured P2P overlay networks Slide 75
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
CAN: Example Yann Busnel75Structured P2P overlay networks Slide 76
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
CAN: Example Yann Busnel76Structured P2P overlay networks Slide 77
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
CAN: routing Yann Busnel77 (a,b) (x,y) (1) a = h x (K) b = h y (K)
y = b node X ::insert(K,V) x = a (2) route(K,V) -> (a,b) (3)
(a,b) stores (K,V) Structured P2P overlay networks Slide 78
UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE
(x,y) CAN: node insertion Yann Busnel78 (1) Bootstrap : discovery
of a contact node already participating to the CAN overlay network
(2) Selection of a random point (p,q) in the space (3) Routing to
(p,q) and discovery of node Y (4) Zone splitting between Y and N
(p,q) Y N N Insertion affects only Y and its immediate neighbours
Structured P2P overlay networks Slide 79 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Routing information
The joining node gets the IP @ of its neighbours from the previous
owner of the zone Set of neighbours of the joining node is a sub-
set of neighbours of the previous owner The previous owner updates
its own list of neighbours The neighbours of the joining node
should also be updated Yann Busnel79Structured P2P overlay networks
Slide 80 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Departure, arrival, maintenance Node departure The
leaving node L must make sure that the zone is taken over A leaving
node hands over explicitly its own zone (and associated database)
to one of its neighbours Failure: detected by periodic messages
from a node to its neighbours (hearteat) containing its own
coordinates and coordinates of its neighbours Yann
Busnel80Structured P2P overlay networks Slide 81 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE CAN:
properties Yann Busnel81 Each node maintains pointers to its
immediate neighbours = 2d O(d) Routing in a N node network Number
of hops in a d-dimension space: In case of failure: selection of an
alternative neighbour Optimizations Multiple dimensions Multiple
reality RTT Measures Zone splitting Locality awareness: landmarks
Structured P2P overlay networks Slide 82 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Failure resilience
Yann Busnel82 destination source Structured P2P overlay networks
Slide 83 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE NANTES
ATLANTIQUE Failure resilience Yann Busnel83Structured P2P overlay
networks Slide 84 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Failure resilience Yann Busnel84 destination
Structured P2P overlay networks Slide 85 UNIVERSIT DE NANTES
LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Routing resilience
Yann Busnel85Structured P2P overlay networks Slide 86 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Failure
resilience Node X::route(D) If (X cannot progress directly towards
D) Check if one neighbour can progress towards the destination If
so, forward the message Yann Busnel86Structured P2P overlay
networks Slide 87 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Multiple dimensions Increasing the number of
dimensions The average path length is improved The number of
neighbours increases linearly with the dimension Enhanced
availability: potentially more nodes available Yann
Busnel87Structured P2P overlay networks Slide 88 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Reality
Multiple independent coordinate spaces Each node is associated to a
different zone in each space (reality): r sets of coordinates
Enhanced availability DHT content can be replicated across
realities Ex: a pointer to a file stored at (x,y,z) is stored on
three nodes responsible of point (x,y,z) in 3 realities Improves
average path length as well: depending on the destination, the most
relevant reality is chosen Yann Busnel88Structured P2P overlay
networks Slide 89 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE RTT measures So far, the metric used to progress
in the space in the path length in the Cartesian space Better
criterion to take into account the underlying topology RTT to each
neighbour Message forwarded to the neighbour for which the ratio
progress/RTT is the best Avoid long hops Yann Busnel89Structured
P2P overlay networks Slide 90 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Summary on structured overlay
Chord, Pastry and Tapestry use a generalized hypercube routing:
prefix matching State maintained: O(Log(N)) Number of routing hops:
O(Log(N)) Proximity routing in Pastry and Tapestry CAN uses
progression in a multidimensional Cartesian space State maintained:
O(D) Number of routing hops: O(N1/D) Proximity routing more
difficult to exploit Yann Busnel90 DHT Functionality = Exact match
interface Structured P2P overlay networks Slide 91 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Summary on
structured overlay Avantages Localization guarante in determistic
time Disadvantages Maintenance cost of DHT Weak autonomous Poor
query language (exact matching) Yann Busnel91Structured P2P overlay
networks Slide 92 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Conclusion Structured architectures ensure
localization of an object among the networkl Contrepart is link to
the maintenance cost and peer autonmoy loss Yann Busnel92Structured
P2P overlay networks Slide 93 UNIVERSIT DE NANTES LABORATOIRE
DINFORMATIQUE DE NANTES ATLANTIQUE Comparison of P2P systems
UnstructuredHybridStructured AutonomyHighMediumPoor
EfficiencyPoorMediumHigh Language expressivity HighMediumPoor Fault
toleranceHighPoorHigh Unstructured P2P/ Hybrid P2P / Structured P2P
Yann Busnel93Structured P2P overlay networks Slide 94 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE Comparison
(2) CONCLUSION : Not one fully satisfied solution ! Poor Medium
High Unstructured Hybrid Structured Autonomous Expressivity
Robustness Localization quality Yann Busnel94Structured P2P overlay
networks Slide 95 UNIVERSIT DE NANTES LABORATOIRE DINFORMATIQUE DE
NANTES ATLANTIQUE Conclusion 3 P2P architectures type considered,
which respect more or less the original paradigm : Unstructured
(respect paradigm) Hybrid (remove equality between peers)
Structured (remove autonomous principle but permit indexation)
Choice of an architecture is made according to: Applicative context
Constraints Challenges Enhance localization process, take into
account higher-level query language, securize interactions Yann
Busnel95Structured P2P overlay networks Slide 96 UNIVERSIT DE
NANTES LABORATOIRE DINFORMATIQUE DE NANTES ATLANTIQUE STRUCTURED
P2P OVERLAY NETWORKS Yann Busnel96Structured P2P overlay networks
Yann Busnel Maitre de confrence Universit de Nantes F2R401b -
Architectures des systmes dynamiques Telecom Bretagne Rennes
