P2P 2 or Not 2 P2P
2 P2P or Not 2 P2P
P2P Not P2P
What questions should a system designer ask to judge weather a P2P solution is appropriate for his particular problem? * a heuristic decision tree
Presented by: Ghada Alnifie
P2P Environments
Self-organizing
nodes organize them selves into a network through a discovery process
Problem Characteristics Axes
Decisions to be considered:
Budget
- Ample or limited? - Likelihood that a unit of service is interesting to many peers - The cost of handling mutually distrusting peers is high - Timeliness and consistency - Solving critical problems may need centralized solution
Resource Relevance to Participants
Symmetric Communication Decentralized Control
peers are considered equals; they both request and offer services there is no central controller that dictates behavior to individual nodes
Trust
Rate of System Change Criticality
1
Candidate Problems
Routing Problems
takes on p2p characteristics when the scale is large enough or when centralization is ruled out - Internet Routing - Ad hoc in Disaster Recovery - Metropolitan-area Cell Phone Forwarding
Candidate Problems
Backup
the process in which a user replicates his files in different media at different locations to increase data availability - Internet Backup - Corporate Backup
Candidate Problems
Distributed Monitoring Monitoring in large distributed systems
- simple (publish/subscribe) - complicated online manipulation (SQL queries) - the basis for an off-line study
Candidate Problems
Data Sharing
peers offer the data to be shared and also search collection to find their interest
- File sharing - Censorship Resistance
2
Candidate Problems
Data dissemination
same as data sharing, with the difference that data is stored to be spread (forwarded) - Usenet - Non-critical Content Distribution
2 P2P or Not 2 P2P
Problem characteristic
!
P2P may be inappropriate
Problem class
Conclusions
The limited budget requirement is the most important motivator, relevance comes second since can be compensated for. Problems that lack these 2 requirements are not appropriate for a P2P solution. Trust between nodes greatly eases P2P deployment.
Second Paper
Exploring the Design Space of Distributed and Peer-to-Peer Systems: Comparing the Web, TRIAD, and Chord/CFS
3
Introduction
This paper compare several distributed and P2P systems by evaluating a key set of architectural decisions:
A family of Distributed systems
WWW Distributed file systems The telephony network P2P systems
*latest addition
- naming, - addressing, - routing, - topology, and - name lookup.
Design Axes of Distributed Systems Content name Host address
what where how
Comparison
WWW, Triad, and Chord/CFS are compared. To show how the different architectural choices impact availability, redundancy, security, and fault tolerance of such systems.
Routing mechanism Network topology Lookup
bindings
links
4
The World Wide Web
Perhaps the most ubiquitous, popular, and successful distributed system. Enables clients to retrieve hyperlinked content. Names: infinite space. Addresses: globally unique IP addresses Routing: a combination of Internet routing protocols Topology: hierarchical, consisting of interconnected autonomous systems and sub-networks within them Lookup: URLs are resolved to IP addresses through the domain name server (DNS)
TRIAD
Defines a content layer that replaces the Webs address-based routing with a name-based routing uses protocol.
TRIAD
Names: the Webs URL namespace for content naming Addresses: a composition of two namespaces: globally unique IP addresses of AS, and locally unique IP addresses within each AS. Routing: uses name-based, BGP-like routing protocol Topology: can be arbitrary consisting of logical links between relay nodes Lookup: unifies lookup and routing
Chord/CFS
Hosts serve as servers, clients, and intermediate routers. Names: a Chord identifier is obtained by hashing Addresses: Addresses are obtained by concatenating a hosts IP address with a small virtual host number, and hashing the result into a 160 bit address. Routing: can be thought of address-based or namebased. Topology: a deterministic function of participating peers addresses Lookup: unifies lookup and routing
5
Summary
How do these different architectural choices impact the performance of a distributed system?
Names and Addresses
1) WWW:
Hierarchical DNS a malicious web server cannot: - register web names randomly - cannot attack another web servers content by duplicating its URL Hierarchical IP makes it difficult for a malicious host to hijack an IP address outside of its allocated range ** although both hierarchal, name space and address space are completely independent in the web
Names and Addresses
2) TRIAD:
Content names are modeled after URLs, this is important for scalability. The ability to create a name in TRIAD is unrestricted. Restrictions on binding rights must be enforced by the routing infrastructure; to date, this issue remains unsolved. Similar to web, individual hosts cannot affect IP assignments. Lookup and routing are unified
Names and Addresses
3) Chord/CFS:
Content name space is flat: - unless the right to insert a name-to-address binding is controlled, any host can cause unbound amounts of effort and storage to be expended across the system. - attacks on a specific victim are possible The set of content names associated with an address is deterministic (possibility for attacks) Lookup and routing are unified
6
Routing, Lookup, and Topolo...