Date post: | 11-Jan-2016 |
Category: |
Documents |
Upload: | phillip-booth |
View: | 219 times |
Download: | 0 times |
Peer-to-Peer Beyond File Sharing:Where are P2P Systems Going?
HotP2P 2009 Rome, May 29 2009
Renato Lo Cigno Tommaso Pecorella, Matteo Sereno, Luca Veltri
Thanks
• To my co-authors– Tommaso, Matteo, Luca
• To the entire Profiles staff, who made this talk possible
http://profiles.disi.unitn.it
• University of Trento – Networking Group • University of Firenze – Telecommunication network Lab• University of Parma – Department of Information engineering• University of Torino – Department of Computer Science• University of Catania – Department of Informatics and Telecommunications• University of Pavia – Electric Communication and Remote Sensing Lab• University of Rome II – Networking Group
Outline• Where we started
– Some bits of retrospective
• Where we are – Applications– Security– Business
• Where are we going– Applications– Expectations– Trends– Security
• Outrageous thoughts– Providers’ based P2P– CrossLayer P2P– P2P & IPv6
Outline
• Where we started– Some bits of retrospective
• Where we are – Applications– Security– Business
• Where are we going– Applications– Expectations– Trends– Security
• Outrageous thoughts– Providers’ based P2P– CrossLayer P2P– P2P & IPv6
(Pre)History Bits
• “Peer-to-Peer” definitions
– A class of systems and applications that use distributed resources (CPU, memory, bandwidth, ...) to implement critical functions
– Communication paradigm where the entities/actors are providers and users at the same time
• Internet was indeed born following a P2P approach
– The first nodes were always client and servers at the same time!!
Initial Studies
• Sharing information at the application level– Search and Retrieval problems DHTs, Gossipping
protocols, ...– Closed communities Rare Contents, Finding
Buddies, ...
• Exploiting unused (CPU) resources for distributed computing– Often not exactly P2P, but “voluntary donations”
• Supporting privacy and anonymous communications
Initial Studies
• P2P recognized as a possible different communication/distributed applications paradigm, contrasted to client/server
• Sharing involves much more than files: – Bandwidth, CPU, Storage, Knowledge, ...
• The key enabling factor is that resource composition is more than the sum of resources – some resources do not scale (try to buy a single
processor 100 times more powerful of your Pentium!)– others are localized (local bandwidth comes for free,
global bandwidth is very costly)
History Bits
• P2P exploded with file sharing – legal problems made it infamous– some service components were centralized
• P2P paradigms is now considered for a bunch of applications– Content Distribution and Distributed Storage– Personal Communications (Instant Messaging, Telephone) – Distributed computing (grid, SETI@home)– Video Distribution (Application Level Multicast) – Gaming– ...
• Depending on the measurement point P2P traffic represents between 50% and 90% of all Internet traffic
Typical Example: Bandwidth Sharing for Massive File
Distribution
Massive File on Fat Server
A Bunch of PCs to Reach
Cooperative Distribution: Intuition
Source server: 100 Mb/sClients: 10 Mb/s1. Antivirus update100,000 clientsFile: 4 MB2. Daily database update1000 clientsFile: 600 MB
Source server: 100 Mb/sClients: 10 Mb/s1. Antivirus update100,000 clientsFile: 4 MB2. Daily database update1000 clientsFile: 600 MB
Client/Server
1. 9h:52m2. 14h:48m1. 9h:52m2. 14h:48m
Cooperative
1. 52s2. 09m:54s1. 52s2. 09m:54s
Cooperative Distribution: Intuition
Client/Server Cooperative
1. 9h:52m2. 14h:48m1. 9h:52m2. 14h:48m
1. 52s2. 09m:54s1. 52s2. 09m:54s
Source server: 100 Mb/sClients: 10 Mb/s1. Antivirus update100,000 clientsFile: 4 MB2. Daily database update1000 clientsFile: 600 MB
Source server: 100 Mb/sClients: 10 Mb/s1. Antivirus update100,000 clientsFile: 4 MB2. Daily database update1000 clientsFile: 600 MB
Cooperative
Client/Server
Outline• Where we started
– Some bits of retrospective
• Where we are – Applications– Security– Business
• Where are we going– Applications– Expectations– Trends– Security
• Outrageous thoughts– Providers’ based P2P– CrossLayer P2P– P2P & IPv6
Background
• Internet has been enormously successful due to its resilience
• Client/server model both as transport paradigms and as business
• Business dominated by
– network providers
– information intermediates (Google, Yahoo, ...)
• Service providers are agonizing ... why??
The Client-Server Doom
• Dominating Paradigm– Traditional business models– Intuitive and well known
deterministic technicalities– Clear entities relationships– But ...
Large cost of ownership (CapEx)
Client-Server Doom: Some (failing) answers• Push specialization and outsourcing as means of
optimization and cost reduction• Exploit scale-economy
– centralized solutions to reduce personnel and competences
– resource sharing between different applications and companies
• Consolidate networking and services – Service Providers want to have networks– Network Providers seek services to sell
What is Pushing P2P today?
• Can the C-S model scale without limits? • How does it react to diminishing revenues because
services are perceived as commodities?• How does information “generation” and
“consumption” affect the C-S model? – The wiki and commons models are shaking traditional
ways of knowledge dissemination
• Do we always need a “server” to provide “services”?• Why does the model of network and service
separation does not take off?
But Why P2P
• Cost: Can offer services much cheaper; especially CapEx is low– Think of TV ...
• Enabler for new applications – Can offer services that would not be possible otherwise (think of
out-of-country TV broadcasting or anonymous services)
• Internet (C-S model) itself is losing its resilience and capacity of evolution
• Powerful end systems that are vastly underutilized– Gigaflops
– 100s of Gigabyte of local disk storage
– Good network connectivity
Available P2P applications
• File sharing, • Voice, Video, Chat, • Backup
– few, not much used
• Versioning – in its infancy
• Computing – what’s beyond BOINC? – is it really P2P??
But also • Anonymous networking (tor) and other applications that
may not exist without the P2P approach
However .... There are Open Problems
• Distributed algorithms are needed for– Service discovery– Overlay management (topology, join, leave, ...) – Application level routing
• Reliability and Dependability– QoS definition and provisioning– Trust management, security, privacy
• Efficiency – Mapping between the overlay and the underlying
resources
Objects: Finding and Managing them
• Representing and Abstracting Information– data representation– DHT algorithms – Protocols and Signaling
• Replicating and Consolidating Information– Minimal duplication – De-correlation and De-localization– Aging information: when is information up-to-date?– Ownership and access: who can
access/modify/delete the information?
Resource Management
• P2P Overlays survive only if the resources underneath are available– Where are they, how to find them, how to guarantee
their availability– Cooperation, not Exploitation
• Game theory may not suffice in finding equilibria in cooperation
• Different applications have different interaction models– machine-to-machine– machine-to-human– human-to-human
Real-Time applications
• Typically human-to-human• Two entirely different system models
– Share only the need of short delays ... and indeed not even these
• Live Streaming– Potentially huge audience– Low Jitter, limited delay (1s—20s)– Structured or Unstructured??
• Conferencing– Small groups– Definitely structured– Very low delays (100s ms), zero jitter
Security & Privacy
• Either neglected (in current applications, not in research!!)
• Or centralized (skype and others) • Or the only goal (anonymous networking)• Security and Privacy still thought as of separate
problems ... or even contrasting• Different attacks possible in P2P
– Routing– Partition– Retrieval– Sybil and Eclipse
Considerations (1)
• P2P has very interesting characteristics– Self-scaling: Use P2P wherever the demand is difficult to
predict and very bursty• Very cost-efficient• Substitution technology for applications that can be
done using the client-server paradigm, such as – file distribution, CDN, video streaming, or backup
– Enabler for new applications such as• Anonymous and Privacy Aware Networking• Reliable multicast• Distributed network monitoring and trouble shooting• IPTV
Where are we?
Considerations (2)
• P2P systems are now mainly scattered pieces of software, without coordination between them and without coordination with the network below
• Merging multiple applications on a single overlay management platform starts to be considered
• Mapping the overlay on the underlay resources is intriguing, but awfully difficult
• Cooperation between the network (IP) level and the overlay may increase performances, but both operators and users were against this idea ... until a little ago
Where are we?
Outline• Where we started
– Some bits of retrospective• Where we are
– Applications– Security– Business
• Where are we going– Applications– Expectations– Trends– Security
• Outrageous thoughts– Providers’ based P2P– CrossLayer P2P– P2P & IPv6
Dominant applications
• Traditional file sharing (BitTorrent & the like)– Still generating the largest traffic– Slowly loosing their “forbidden” appeal, exchanging all
possible information– Probably the only “pure” P2P systems so far (at least
some of them)
• Voice & Chat– Lack of standards hamper the evolution
• Video and TV– If not live (VoD) falls in file sharing class– P2P-TV is exploding, with all the promises and
concerns– Conferencing is lagging behind (as usual)
Network-Application Cooperation (NAC)
• Some BitTorrent dialects are already considering form of network awareness to improve performance
• An application which wisely uses resources is cheaper to run
• Operators are starting to consider P2P as a resource, not a foe – after all if it was not for P2P they would not have sold large bandwidth access
NAC: What can ISP do?
• Application Level Redirection
• Application Cashing
• Introduce “network peers” in strategic points
• Manipulate routing
• Manipulate resource availability to force local overlays
Multi-Application Platforms
• P2P is fundamentally re-inventing the communication level every new application: inefficient!!!
• We need to understand P2P as a novel paradigm for communication/computing and provide high-level models that – Enable the definition of general purpose P2P
platforms– Guarantee interoperability
• P2P support as a novel distributed OS
P2P Standardization
• Should we standardize P2P? – Signaling: IETF is working (SIPpeer, ALTO, ...) – Transport: maybe is time to go beyond UDP/TCP ...
• Is PTP (P2P Transport Protocol) useful/needed??
• What does it mean– Protocols only? – APIs? – End devices and OS that are P2P aware
• Segregating local and shared data• Managing the resource/information sharing policies
• Business needs standards (or not?)
Vulnerability & Resilience
• Byzantine resistant systems are available– Based on approximate Nash Equilibria– Complex– Is this what we need?
• P2P systems are resilient – To random mistakes/failures– To (many) external attacks
• Still they are quite unreliable– Let’s ConfCall with Skype ... and fall back to PSTN if it
does not work ... – Why?
Privacy
• Little considered in general• Sociology studies show that people are willing to
trade privacy (when they know what it means) for commodity
• Little understanding of the implication of linkage in general
• Once more TV example– Standard Broadcasting– Server-Based IPTV– P2P Options
• Server-based • Anonymous, DHT based
Outline• Where we started
– Some bits of retrospective
• Where we are – Applications– Security– Business
• Where are we going– Applications– Expectations– Trends– Security
• Outrageous thoughts– Providers’ based P2P– CrossLayer P2P– P2P & IPv6
Providers & P2P• Server-assisted P2P is becoming a promising
networking (and business) model at least in TV (but BitTorrent dialects AS-aware exists as we said)– PPLive, UUsee, ...
ISPISP
ISPISP
ISPISP
ISPISP
ISPISPISPISP
ISPISP
Degradedservice
Providers & P2P• ISP-Assisted is a different model
– Do you trust more ISPs or Service Providers?– Set-Top boxes can play the game
ISPISP
ISPISP
ISPISP
ISPISPISPISP
ISPISP
Cross layering
• Anything farther then Application –Level Networking??
• tor is an example of [cross-layer]-1
But• Media-aware schedulers• Network Aware Topologies• TCP-friendly P2P
are all examples of “cross-layering”
IPv6 & P2P
• Restores end-to-end semantics (no NATs, ...)
But• Somebody (Cornell) is implementing IPv6
routers as a P2P overlay on top of IPv4!!!• Current P2P applications are relaying on IPv4
details• Maybe IPv6 addresses can be used as a
common platform for P2P addressing ... leaving IPv4 where it is