Peer-to-Peer System
< key words>
1. Cache
2. Happy Eyeball = request/access navigation
3. Contents (versus Stream)
What is P2P
•
Apply the technical frameworks, that is applied
to inside a computer so as to improve the
processing performance, are applied to the
distributed computer network environment.
•
In particular, the followings are key components
for P2P system;
1. Cache (and mirroring)
2. Redirection of data flow, e.g., DMA
3. Abstraction of objects, to handle by unique and single method.
Something new in the Internet
1. Multi-processor computer architecture for
distributed processing, aka functional
distribution, geographical distribution
2. Cache and Proxy
CDN(Contents Delivery Network)
3. Virtualization
Background of P2P;
Unawared technical evolution ?
•
Two of drastic cost down in ICT
– Copying
– Recording and saving
•
However, logistics of information may not be
aware of this evolution……
– Have BitTorrent or Joost been awared ?
2006/10/11 同 時視 聴 者数 と送 出 トラ フィ ック の 推移 グラ フ 17:01 17:17 17:33 17:49 18:05 18:21 18:37 18:53 19:09 19:25 19:41 19:57 20:13 20:29 20:45 21:01 21:17 21:33 21:49 22:05 22:21 22:37 22:53 23:09 23:25 23:41 時間 B B B 同時 視聴者 数[単 位：人] セン ター トラ フ ィック [ 単位： Mbps] (HMS x7 2 総計) 中継配信開始⇒ サーバーの送信トラフィック 同時視聴者数 ? 中継配信終了 37.3Gbps (=768Kbpsx48,454) With legacy Unicast-based multicast Time Start Program Bandwidth at server Number of clients Game-set 6.97Gbps (=37.3Gx18.7%) October 11, 2006.
Play-off game of professional baseball, Hokkaido Nihon Ham Fighters versus Fukuoka Softbank Hawks.
Peer-to-Peer overlay multicasting service by professional ISP, BB-TV! by SoftBank over ADSL network
6
Peer-to-Peer Overlay Networking
•
First generation
– Napstar, WinMX
directory server + Pee-to-Peer connection (*) similar to SIP and NGN
•
Second generation
– Gnutella
Server-less pure peer-to-peer (*) have a scaling problem…..
•
Third generation
– Freenet, Winny
7
Napster
server PC PC Store contents File name/username Request to connect Provide contents Store contents IP Addrss, Port, Path (URI) Index of contents8
Gnutella
Peers/ neighbors
9
Gnutella
：
Discovery
Got it !
Peer/neighbor
Have a retrieved contents data A B C D E F G
10
Gnutella
：
Transmission
Peer/neighbor
Node has a target contents data A B C D E F G data
11
Freenet
：
put a file
自分のピア
※HTL（TTL）=2
•GUID key is broadcast
Contents holder Source data Network Hashvalue Hash value Hash value
12
Freenet; file transfer
自分のピア 求めるデータのあるピア A B C D E F G data data
図9-5 Winny/SKYPEにおける階層的トポロジー構造の概念図 Super-Nodes
cache/
Distributed file storage in P2P system
•
Napster
– Separate Index and Storage
• DMA operation by the introduction of out-band signaling
• Reducing the copy overhead using the pointer
•
Freenet/Winny
– Separate file name and it’s location information
• Operation like virtual memory system
– Anonymity of file profile
– Introduction of network cache
What has Peer-to-Peer introduced ?
1. {networked} Cache and Proxy
2. {networked} DMA (Direct Memory Access)
3. {networked} Virtual memory system (by DHT)
– Separate contents handler(identifier) and real storeing address
– Access heterogeneous and multiple device with single {virtual} interface
4. Abstraction of contents by number (by DHT)
Cache memory in computer
Main memory
High speed processing Lower speed processing
① Requesting “read” data
② Responding “send” data
(*) ②is so slow idle clocking at CPU….
Introduction of cache memory system
Main memory
<<High Speed>> ① Requesting <<Lower Speed>>
“read” data ②Responding “send” data Cache memory Data A Data A <<High Speed>>
(*) Send the data by the previous and old request
What has Peer-to-Peer introduced ?
1. {networked} Cache and Proxy
2. {networked} DMA (Direct Memory Access)
3. {networked} Virtual memory system (by DHT)
– Separate contents handler(identifier) and real storeing address
– Access heterogeneous and multiple device with single {virtual} interface
4. Abstraction of contents by number (by DHT)
DMA; Direct Memory Access
main memory I/O device DMA controller① Ordering input/output data
② “read” and “write” data
③ Acknowledge of
completion of operation (interruption)
During the DMA operation, CPU can not access the main memory
Signaling system
•
Out-band signaling
– Separate the Data-Plane and Control-Plane)
– Example; Telephone, MP3 player, TV broadcasting
•
In-band signaling
– Share and use the same resource by Data-Plane and by Control-Plane
What has Peer-to-Peer introduced ?
1. {networked} Cache and Proxy
2. {networked} DMA (Direct Memory Access)
3. {networked} Virtual memory system (by DHT)
– Separate contents handler(identifier) and real
storeing address
– Access heterogeneous and multiple device with
single {virtual} interface
4. Abstraction of contents by number (by DHT)
Hash Value 0 A B C D E F G H
Store the files, whose Hash value is “0”～”A”
Store of “A”～”B” Store of “B”～”C” Store of “C”～”D” Store of “D”～”E” Store of “E”～”F” Store of “F”～”G” Store of “G”～”H”
Store of larger than “H”
How the real network looks like ?
1. Covering our “Earth” with high speed network
2. Design the “earth” scale computer system
3. Impossible to accommodate earth with single
technology
4. Investment and operation is always autonomous
5. We have large area, where we could not wire
6. We have large area where, even, wireless would
be hard to use
7. Uni-Directional Digital Link
2001 InternetTraffic
2005 InternetTraffic
Increase of Traffic at BRICS, and The Internet is surrounding the “Earth”.Contents delivery paths
Issue of global Web Service
ISP iDC ISP コンテンツプロバイダ Internet ISP IX iDC ISP 利用者 IX Contents Contents Contents
CDN as scaling mechanism
•
How to come up with the gap between
Moore’s law and Coffman’s observation
– Reverse proxy
– Mirroring
•
Improve end-to-end delay/latency for
global service
ISP iDC ISP コンテンツプロバイダ Internet ISP IX iDC ISP 利用者 IX キャッシュ キャッシュ キャッシュ Contents Contents Contents Contents
CDS
・
CDNSD
4 5 0
EN TER PRISE
Sun DRIVENULTRASPARC
SD NetApp F210 NetworkAppliance Cache, mirror Original Contents Server SD PROFESSIONAL WORKSTATION W PROFESSIONAL WORKSTATION V70 SD USER Text information is From Original node Rich Contents are cached Aka, reverse cache
DNS-based CDN
aka, redirection, like napstar
ISP Network Cache DNS query DNS response Cache GSLB DNS Heartbeat Request navigation
Reverse proxy + URL rewriting
Internet Akamaizer Origin server SLB with URL- rewriting & DNS GET index.html GET foo.gifＨＴＴＰ
forward Cache
Internet
[note]
A=Residential customers A1: Broadband users A2: Non-Broadband users B=Non-residential customers
B1: Traffic exchanged through IXes B2: Traffic exchanged private peerings B3: Oversea traffic
33
(1) Large contribution by residential customer traffic, small contribution by corporate networks
(2) Total traffic volume in Japan (2009)
Internet Home : 1.36 Tbps
International
Incoming traffic is exception; steep increase
Non-residential customer traffic Total residential customer traffic
34
<< Downloading traffic : Network Customer >> << Uploading traffic : Customer Network >>
Peak time seems to shift toward earlier time-zone, i.e.,
premier time for TV
Large volume of
ambient (background) traffic
Weekly Broadband customer traffic for three years
Large volume of
ambient (background) traffic
Weekly International Traffic Volume for three years
<< Outgoing traffic ; Japan Overseas >> << Incoming traffic ; Japan Overseas >>
1. Incoming(import) traffic dominates
outgoing(export) traffic 2. Large variation in
incoming, i.e., interactive traffic
<< Current traffic volume >>
Variation amplitude increases significantly. Due to interactive
traffic ?
Mirroring or caching in Japan may improve this situation
Daily traffic volume per customer
Cumulative distribution for total traffic per customer
10% of users exchange more than 500MBytes
per day
1% (1/100) of users exchange more than
Recent discussion with ISPes;
Which is a headache for ISP,
P2P or Interactive Video ?
P2P wave seems to have small amplitude
Log {Total Traffic per Customer}
User Density
Interactive Traffic, e.g., VoD/Streaming
P2P traffic
User Density versus Total Traffic Volume per Customer
Log {Total Traffic per Customer}
User Density
Interactive Traffic, e.g., VoD/Streaming
User Density versus Total Traffic Volume per Customer
Mode
(=Peak density)
Average traffic volume per customer
Log {Total Traffic per Customer}
User Density
P2P traffic
User Density versus Total Traffic Volume per Customer
Downloads vs Uploads
Challenge on P2P deployed in ISP
•
Traffic engineering on P2P system
–
Intra-domain
P4P ?
–
Inter-domain
new routing
matrix
Collaborating with P2P Promotion Council http://www.fmmc.or.jp/p2p_web/index.html
Challenge on P2P deployed in ISP
•
Traffic engineering on P2P system
–
Intra-domain
P4P ?
–
Inter-domain
new routing
matrix
Collaborating with P2P Promotion Council http://www.fmmc.or.jp/p2p_web/index.html
Distinguish;
Proxy, Overlay, and Gateway
• Overlay : building “logical” and “virtual” network over some physical network (e.g., IP network)
hard to optimize the system operation
• Proxy: intermediate node snoop the “Transparent” communication to behave optimized responses to improve the system efficiency
effectiveness depends on system behavior
• Gateway: Translate different languages(protocols)
less effective….. Need huge state-full machine to inter-network large scaled two inter-networks
