Upload
miller
View
41
Download
5
Embed Size (px)
DESCRIPTION
Towards Efficient Simulation of Large Scale P2P Networks. Tobias Hoßfeld. ITG-Fachgruppe 5.2.1 “Cooperating and Scalable Networks” Aachen, Ericsson, J. Sachs, 04.05.2006. Cartography of P2P Architectures. user-oriented domain. decentralized. X. Gnutella. operator-centric domain. X. - PowerPoint PPT Presentation
Citation preview
University of WürzburgInformatik III (Distributed Systems)Prof. Dr. P. Tran-Gia
www3.informatik.uni-wuerzburg.de
Towards Efficient Simulation of Large Scale P2P
NetworksTobias Hoßfeld
ITG-Fachgruppe 5.2.1“Cooperating and Scalable Networks”
Aachen, Ericsson, J. Sachs, 04.05.2006
2
University of Würzburg
Distributed Systems
T. Hoßfeld
Cartography of P2P Architectures
Two control functions in P2P systems
Resource mediationwhere are files located
Resource access controlwho may download a file and when
Mapping of P2P architectures into architectural space pure P2P hybrid P2P classic client/server
Identification of control objectives
Client/Server
PureP2P
X eDonkey
X BitTorrent
X Gnutella
X Chord
resource access controlcentralized decentralized
res
ou
rce
me
dia
tio
nce
ntra
lized
dece
ntra
lized
P2P Cartography
operator-centricdomain
user-oriented domain
hybrid P2P
3
University of Würzburg
Distributed Systems
T. Hoßfeld
Structured P2P-Network (Chord)
InformationProvider
InformationMediator
InformationSeeker
Peer-to-Peer Architectures
Unstructured P2P-Network (Gnutella)
InformationProvider
InformationSeeker
BitTorrent
TrackerWeb-Server
InformationMediator
Hybrid P2P-Network (eDonkey)
Index
IndexInformationMediator
4
University of Würzburg
Distributed Systems
T. Hoßfeld
Basic Functions of P2P Networks
join() leave()new nodeknows
one peer
bootstrapserver
contacts
position in the P2P network according to the structure: self-organization
node failures detected by periodical updates or not answered requests: resilience
peer informs its “neighbors” and bootstrap server
bootstrapserver
5
University of Würzburg
Distributed Systems
T. Hoßfeld
Basic Functions of P2P Networks (contd.)
insert(key,data) retrieve(key)
peer x wants to insert (key,data); using DHTs: key = hash(data)
x
key routes to peer y which stores datay
structure assigns responsibility for data based on hash function: load balancing
y responsible for (key,data)
peer x searches for key
y has (key,data)
xand asks its neighborswhich redirect requests
y sends data to x
performance improved by using shortcuts according to the known structure for a given key:scalability
6
University of Würzburg
Distributed Systems
T. Hoßfeld
Main feature is multiple source download. Peers issue several download requests for the same file to
multiple providing peers in parallel. Providing peers serve the requesting peers simultaneously.
Basic Function of P2P Content Distribution
index server #1
index server #2
downloading peer 1
providingpeer
providingpeer
providingpeer
downloading peer 2downloading
and providingpeer 1
After successfully downloading a whole
chunk, it is provided to other peers.
7
University of Würzburg
Distributed Systems
T. Hoßfeld
Features of P2P Systems and Their Implications
Usually a large number of participating peers Large-scale: a lot of nodes and even higher number of resources
need to be simulated
Peers may arbitrary join or leave Highly dynamic: a lot of user created event (due to churn, i.e.
peers joining and leaving arbitrarily, as key feature of P2P systems)
Cooperative working of peers and robust systems Complexity:
one event can cause a large number of events at other peers, i.e. system events, due to cooperation among peers
additionally periodic or provisional systems event to cope with the self-organization of p2p systems
Target of Workshop:Focus on large-scale P2P networks in order to consider key characteristics (e.g. regarding churn for 100 peers reasonable?)
8
University of Würzburg
Distributed Systems
T. Hoßfeld
Approach For Simulating Large-Scale P2P Systems
System state has to be stored at simulation machine requires efficient data structures (e.g. calendar queue)
How to model in order to reduce the number of events? Resource mediation might not require to model bandwidth,
only signalling delay Resource exchange might not require to model delay if large
contents are exchanged; requires modelling of bandwidth other performance influence factors: packet loss, moving
users, … appropriate abstractions & models for investigated application
Clustering of peers to user groups... might allow parallel simulation of clusters
9
University of Würzburg
Distributed Systems
T. Hoßfeld
Workshop in Würzburg
Efficient Data Structures Andreas Binzenhöfer, Calendar Queue and Event Design
Algorithms Jens Oberender, Modelling Resource Fragmentation
Abstractions and Models Kolja Eger, Packet-based Simulation Gerald Kunzmann, Signaling in Voice/Video over IP Systems Daniel Schlosser, Tobias Hoßfeld, Periodic and Market-Based
Bandwidth Allocation
Parallel Simulation Ivan Dedinski, Parallel Discrete Event Simulation
10
University of Würzburg
Distributed Systems
T. Hoßfeld
Talks Today
abstract
detailed
abstract
detailedunoptimized
adapted
Modeling Resource Access Control
Modeling Resource Mediation
Co
mp
ac
t
D
ata
Str
uc
ture
s
TobiasHoßfeld
KoljaEger
GeraldKunzmann
AndreasBinzenhöfer Hier könnte Ihr
Name stehen !
University of WürzburgInformatik III (Distributed Systems)Prof. Dr. P. Tran-Gia
www3.informatik.uni-wuerzburg.de
Periodic and Market-Based Bandwidth Allocation in
eDonkey Networks
Tobias Hoßfeld, Daniel Schlosser
12
University of Würzburg
Distributed Systems
T. Hoßfeld
Measurements of eDonkey Traffic
Case-by-case measurements of eDonkey file-sharing application in public GPRS/UMTS network
Multiple source download via GPRS
0 5 10 15 20 25 300
5
10
15
20
25
30
35
40
45
time [min]
thro
ug
hp
ut
[kb
ps
]
downloading peer (mobile)sharing peer #1 (mobile)sharing peer #2 (fixed)
13
University of Würzburg
Distributed Systems
T. Hoßfeld
eDonkey Data Exchange via UMTS
UMTS upload restricts throughput
UMTS download restricts throughput
0 2 4 6 8 100
1
2
3
4
5
time [min]
da
ta [
MB
]
fixed downloading peerfixed sharing peer #1mobile sharing peer #2
0 5 10 15 20 25 300
1
2
3
4
5
time [min]
da
ta [
MB
]
mobile downloading peerfixed sharing peer #1mobile sharing peer #2
Max-min-fair share of available bandwidth is observed How to model the bandwidth allocation of fair-share P2P file-sharing
applications?
14
University of Würzburg
Distributed Systems
T. Hoßfeld
Simulation of Fair-Share Bandwidth Allocation
Events which influence the bandwidth allocation are that a peer… starts the download of a file finishs a download goes offline while downloading continues downloading a file after joining the network again
We consider eDonkey-like file-sharing networks
Aim: Modeling of bandwidth allocation in fair-share networks
tEvents
tComputation ofallocated bandwidth
t t t t t t t t t t t
15
University of Würzburg
Distributed Systems
T. Hoßfeld
Stream-based or packet-based approach?
TCP can be neglected if conditions are fulfilled (540 kB blocks) Signaling vs. data exchange: RTT vs. bandwidth
0 50 100 150 200 250 300 350 4000
50
100
150
200
250
300
350
400
downlink bandwidth [kbps]
go
od
pu
t [k
bp
s]
RTT = 100msRTT = 200msRTT = 300ms
16
University of Würzburg
Distributed Systems
T. Hoßfeld
What means fair-share?
Peer 311 kbps
Peer 23 kbps
Peer 540 kbps
Peer 420 kbps
uploadingPeer 140 kbps
3 kbps11 kbps
13 kbps 13 kbps
All peers get the same bandwidth If a peer cannot consume completely the allocated
bandwidth, the surplus is distributed among the remaining peers
17
University of Würzburg
Distributed Systems
T. Hoßfeld
Periodic Bandwidth Allocation
Peer 1
Peer 2
Peer 5
Peer 3
Peer 4
For each t, for each peer: compute bandwidth allocation
Allocated bandwidth can be
overbooked or underbooked
18
University of Würzburg
Distributed Systems
T. Hoßfeld
Market-Based Bandwidth Allocation
For each event, consider affected, i.e. connected, peers
All affected peers make a bid
Strategy If there are no other bids, propose
x = not allocated bandwidth / #peers If minimal bid y of all affected peers is smaller than x, then
keep bid y and compute x‘ If all bids are larger than x, then bid x on these
connections
Finish: If lower bid of a connection is the minimal bid of a peer and is repeated
19
University of Würzburg
Distributed Systems
T. Hoßfeld
Market-Based Bandwidth Allocation
Uploadingnetwork links
Downloadingnetwork links
NL8: 40kbps
NL1: 10kbps
NL2: 80kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
NL0: 10kbps3.333
5
20
15
10
20
5
10
40Initial bid: x = BW / #peers
20
University of Würzburg
Distributed Systems
T. Hoßfeld
25
MBBA Example
Uploadingnetwork links
Downloadingnetwork links
NL8: 40kbps
NL1: 10kbps
NL2: 80kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
NL0: 10kbps3.333
5 10
20
15
20
5
10
40
3.333
26.667
5 3.3336.667
25
25
If minimal bid y of all connected peers holds y<x, then set bid y and compute x‘ x = BW / #peers.
If all y<x, keep x.
21
University of Würzburg
Distributed Systems
T. Hoßfeld
3.333
MBBA Example
25
Uploadingnetwork links
Downloadingnetwork links
NL8: 40kbps
NL1: 10kbps
NL2: 80kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
NL0: 10kbps3.333
5 10
20
10
40
26.667
5 3.3336.667
25
25
Finish: If lower bid of a connection is the minimal of a peer and is repeated
22
University of Würzburg
Distributed Systems
T. Hoßfeld
MBBA Example
5
25
2525
Uploadingnetwork links
Downloadingnetwork links
NL8: 40kbps
NL1: 10kbps
NL2: 80kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
NL0: 10kbps
3.333
10
40
26.667
6.667
5
6.66724.444
24.44424.444
Finish: If lower bid of a connection is the minimal of a peer and is repeated
Minimal bid y<x, set y and compute x‘
23
University of Würzburg
Distributed Systems
T. Hoßfeld
24.44424.444
24.444
MBBA Example
Uploadingnetwork links
Downloadingnetwork links
NL8: 40kbps
NL1: 10kbps
NL2: 80kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
NL0: 10kbps
3.333
10
40
26.6675
6.667
10
31.667
31.667
Finish: If lower bid of a connection is the minimal of a peer and is repeated
If all y<x, keep x.
24
University of Würzburg
Distributed Systems
T. Hoßfeld
31.667
MBBA Example
Uploadingnetwork links
Downloadingnetwork links
NL8: 40kbps
NL1: 10kbps
NL2: 80kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
NL0: 10kbps
3.333
10
40
26.6675
6.667
36.666
Finish: If lower bid of a connection is the minimal of a peer and is repeated
25
University of Würzburg
Distributed Systems
T. Hoßfeld
MBBA Example
Uploadingnetwork links
Downloadingnetwork links
NL1: 10kbps
NL2: 80kbps
NL0: 10kbps
3.333
10
26.6675
6.667
36.666
NL8: 40kbps
NL3: 40kbps
NL4: 30kbps
NL5: 10kbps
NL6: 10kbps
NL7: 10kbps
26
University of Würzburg
Distributed Systems
T. Hoßfeld
Comparison PBA vs. MBBA
Exchange of small files
27
University of Würzburg
Distributed Systems
T. Hoßfeld
Comparison PBA vs. MBBA
Exchange of large files
MBBA: computes and allocates immediately fair-share bandwidth -> in real systems this requires some time
PBA: bandwidth overbooked or underbooked for Δt -> in next step bandwidth is adapted
28
University of Würzburg
Distributed Systems
T. Hoßfeld
Conclusion
For P2P content distribution networks, like eDonkey, resource access control is crucial point
Fair-share bandwidth allocation has to be modeled
We have proposed two stream-based approaches which are valid in the considered scenarios Periodic bandwidth allocation PBA Market-based bandwidth allocation MBBA
Depending on the number of events influencing the resource access control PBA or MBBA has to be preferred