P2P Content Distribution - uvigo tv

P2P Content DistributionPablo Rodriguez

Christos Gkantsidis

2

Traditional Content Distribution

Often, large content needs to be distributed to millions of clients:

• Currently:

• Huge server farms

• Infrastructure-based solutions (e.g. Akamai)

slow, expensive, non scalable

Server Farm

3

Content Distribution Evolution

Hype

Realism

Growth

Caching

IP Multicast

CDNs

Akamai

Enterprise

CDNs

Layer-7 Switches

Satellite CDNs

P2P

1999

2000

2001

2002

2003

2004

Disappointment

4

P2P Content Distribution

5

P2P Content DistributionDesktop PCs can help each other!

• Clients become new servers

• Capacity increases with the number of clients

• Limitless scalability and fast speeds at extremely low cost!!

Server Farm

10

100

1000

10000

100000

1000000

10000000

0 7 14 21 28 36 43 50 57

Time (sec)

Num

ber o

f Clie

nts

Ser

ved

Cooperative

Client/Server

4 MB file. Server 100 Mbps. Client 1 Mbps

6

Examples• Updates/Critical Patches (combat virus/worm propagation)

– Adding more servers and egress capacity to absorb pick load is quite expensive

– Alternative solution is to artificially delay clients» Patches do not arrive on-time

• Software Distribution– BitTorrent: successfully distributed 1.77GB Redhat 9

• PodCasting

• Group Information Sharing

• Enterprise content distribution

7

P2P Content Distribution

• Benefits:– Dramatically improves speed– Limitless scalability– Minimum server requirements– Very cheap

• Challenges:– Requires incentives for cooperation– Hard to ensure end2end full connectivity– Security– Manageability– Lack of locality increases transit costs for ISPs– Asymmetric links (traffic engineering)– Variable bandwidth, peers come and go– Need for more sophisticated distribution algorithms

8

P2P Swarming• File is divided into many small pieces for distribution

• Clients request different pieces from the server or from other clients

• Clients become servers for those pieces downloaded

• When all pieces are downloaded, clients can re-construct the whole file

1 2 65

Server

3 4

1 5 6 2 4

1 2 3 4 5 6

3

[Rodriguez, Biersack, Infocom’00]

9

1 2 65

The Challenge

Server

3 4

1 5 6 2 4

1 2 3 4 5 6

3

If there are many users,deciding which is the best piece to download can be very hard!!⇒ Incorrect decisions result in low

throughput, nodes not able to finish, bandwidth wasted, etc.

Solutions that require to have full knowledge of who has what are non-scalable

10

Avalanche:Improving BitTorrent through Coding Techniques

11

Goal

• Provide a very fast and robust P2P file distribution solution

•Current problems in existing P2P solutions:•Rare-blocks are hard to obtain•Tit-for-tat incentive mechanisms decrease speeds•Arrival of new users slows down old users•Heterogeneous nodes do not interact well•Same information travels repeatedly over bottleneck links•Too much dependency from seeds•Sudden departures can prevent peers from finishing

12

Source

The Problem of Efficient Scheduling of Information

Node A Node B

Block 1 Block 2

Node C

Block 1

Block 1, or 2, or 1⊕2?

13

The Avalanche Magic

• To solve problems of existing P2P file distribution solutions, Avalanche uses special encoding algorithms

• Each encoded piece has the “DNA” of all pieces in the file.=> A given encoded piece can be used by any peer in place of any piece

• Encoded pieces are created using linear equations that involve all pieces in the file

• Reconstructing the file requires collecting enough encoded pieces and solving the set of mathematical equations

14

Coding in general• Assume file: F = [x1 x2], where xi is a block.

• Define code Ei(ai,1, ai,2) = ai,1*x1+ ai,2*x2, where ai,1, ai,2 are numbers.

• “Infinite” number of Ei’s.

• Any two linearly independent Ei(ai,1, ai,2) can recover [x1 x2]. – Similar as solving a system of linear equations.

• Operations in finite fields [such as GF(216)].

15

Avalanche Coding

B1 B2 Bn

Server

α1 α2

Client A

β1 β2 βn

E1 E2

Client B

ω1 ω2

E3

[Chou et al., ’03]

• Content is encoded at the server

• Clients can produce new encoded packets out of partial files

αn

File

16

Avalanche Robustness

100 150 200 250 3000

50

100

150

200

250

300

350

400

450

500

Time

# of

Pee

rs F

inis

hed

NCFECLR

If server suddenly goes down (after serving the ful l file one), all Avalanche users are able to complete the download. Only 10% of BitT orrent-like users are able to complete.

17

Avalanche Download Time

Fin

ish

Tim

es

Nodes (sorted by order of arrival)

0 50 100 150 200 250 300 350 4000

50

100

150

200

250

300

NCRandom

Avalanche

BitTorrent

BitTorrent peers not yet finished

=> Much lower and predictable download times

18

No need for nodes to stay around…

• With Avalanche, there is no need for nodes to stay after they finish the download to help other nodes (the performance remains unchanged)

Nodes stay for ever

Nodes leave immediately

Nodes (sorted by order of arrival)

Fin

ish

Tim

es

19

Minimum Server Requirements

Less than half the server requirements of BitTorrent-like systems

50 100 150 2000

20

40

60

80

100

120

140

Time

Ser

ver

Load

NCFECSimple

20

Decoding Performance

Avalanche trades-off better speeds and less server

load for more processing power at each node

3m 38 sec100200

2m 21 sec100100

37 sec10050

5 sec10010

TimeBlocksFile Size (MB)

Note: Pentium III, 650MHz, 512MB RAM.

Decoding time is less than 4% of the total download

21

Summary

•Adding resources in an arbitrary fashion is not efficient or cost effective

•We are witnessing a new Revolution•P2P can be used to provide hugely scalable, fast content distribution at very low cost