1. 09/13/04 CDA 6506 Network Architecture and Client/Server Computing Peer-to-Peer Computing and Content Distribution Networks by Zornitza Genova Prodanoff

2. ZGP002 Lecture Outline Client/Server Model Peer-to-Peer Model Content Distribution Networks

3. ZGP003 The client (process) Runs on a desktop Makes requests for services Executes a portion of the application code The server (process) Executes a set of functionally related services Never initiates a message exchange with a client Implements a queue to hold client requests as they arrive Request-reply protocol: client sends requests and server replies Queue of client requests at the server Server throughput is the number of requests served per unit time Client/Server Communication

4. ZGP004 Server Types: File servers Database servers Application servers Groupware servers Object servers Web servers Software servers Client/Server Communication (continued)

5. ZGP005 Hypertext Markup Language (HTML) −Integrate images and other multimedia objects, as well as software applications Application layer protocol Runs on top of TCP/IP C/S Example: Hyper Text Transfer Protocol (HTTP)

6. ZGP006 HTTP 1.0 vs. HTTP 1.1 HTTP 1.0 −Allows for embedded documents in the HTML file −The client browser opens a separate connection with the server for each requested file: html, image, etc. −Client does not need to wait for connection termination before another connection request is sent −Images are downloaded transparent to user: HTTP response, containing the URL for the image is parsed before a new connection request is made C/S Example: Hyper Text Transfer Protocol (HTTP)

7. ZGP007 HTTP 1.1 −Allows for embedded documents in the HTML file −“Persistent connections”: The client browser opens a (single) connection with the server for the html file. Then, embedded documents are requested within this same TCP connection −Pipelined requests: multiple requests are sent without waiting for response (ACK) −Pipelined HTTP 1.1 outperforms parallel connection based HTTP 1.0 C/S Example: Hyper Text Transfer Protocol (HTTP)

8. ZGP008

9. ZGP009 Peer-to-Peer Model Problems with C/S model −Server seen as a central node: − Scalability − Reliability Peer-to-Peer model: a collaborative and distributed approach to computing A study conducted at a major company with 10,000 workstations −75% of CPU cycles in the workstations or network nodes such as routers are idle −50% of disk space is free

10. ZGP0010 Peer-to-Peer Model (continued) 10 billion MHz & 10,000 TB not utilized at the edges of the Internet (see openP2P.com) The size of the networks and the complexity/requirements from the protocols steadily increase Bandwidth consumption attributed to popular file sharing applications reaches 60% of the total Internet traffic Must be able to locate the resources efficiently

11. ZGP0011 Peer-to-Peer model: sharing of resources available at the edges of the Internet In a P2P system, distributed computing nodes of equal roles or capabilities exchange information directly with each other Resources could be content, storage, CPU-cycles, bandwidth P2P paradigm has many plausible characteristics: −Scalability −No centralized authority, −Robustness −Cooperation, sharing −Anonymity Peer-to-Peer Model (continued)

13. ZGP0013 Routing Indexes No-indexing (Gnutella) Centralized index (Napster) Distributed index (FreeNet) Peer-to-Peer Model (continued)

14. ZGP0014 Advantages of Peer-to-Peer model: Performance gain with much less cost Self-organization Load balancing Adaptation Fault-tolerance Mobile devices Peer-to-Peer Model (continued)

15. ZGP0015 Open problems: Locating and addressing Caching of passing through objects How to identify an Object Network usage and participation Peer-to-Peer Model (continued)

16. ZGP0016 Queries propagate across nodes, each node does some processing and uses resources Bandwidth: The average over a set of representative queries of the aggregate BW consumed (in bytes) over each edge on behalf of the query Processing: The average over a set of representative queries of the aggregate processing power consumed at each node on behalf of the query Peer-to-Peer Model (continued)

17. ZGP0017 Gnutella: BFS technique is used with depth limit of D, where D= TTL of the message. At all levels < D query is processed by each node and results are sent to source and at level D query is dropped. Freenet: uses DFS with depth limit D. Each node forwards the query to a single neighbor and waits for a definite response from the neighbor before forwarding the query to another neighbor (if the query was not satisfied), or forwarding the results back to the query source (if query was satisfied). Peer-to-Peer Model (continued)

18. Next generation Internet - Content Distribution Networks - A CDN is an overlay network on the Internet - A CDN co-locates content throughout the world CDNs are of a great commercial and research interest - $15 million in NSF funding for Web services research - Akamai is one major CDN provider Content Distribution Networks ZGP0018

19. Transparent cache Origin site Reverse cache Distributed server Internet Clients Proxy cache http://www.some.com/page http://334.249.2.8/page Global content distribution in a CDN http://214.29.2.15/page ZGP0019 Content Distribution Networks

20. HTTP redirection in a CDN (1) HTTP request and redirect (2)HTTP re-request and response Origin site Reverse cache Distributed server Clients Proxy cache (1) (2) URL router ZGP0020 Content Distribution Networks

21. Caching: reduces both Internet traffic and server load implemented – local to the client – as a caching hierarchy in the Internet – at the server site content information best shared as compressed digests update interval - seconds to minutes in length ZGP0021 Content Distribution Networks

22. Selection of the “best” server with satisfactory response time to the user and minimized network traffic. (Selection criteria: server load, server contents, and network path characteristics) ZGP0022 Content Distribution Networks

23. Scaling from local to distributed cluster: internet connections constrained by bandwidth and delay –state information shared with less frequency –state information smaller in size “herd effect” –chose least-loaded server –unbalanced and oscillating loads. ZGP0023 Content Distribution Networks

24. One armed URL router HTTP requests and redirects Network links Layer 3 switch Architecture of a new URL router URL 1Loc 1 (state), loc 2 (state), … loc M 1 (state) URL 2 Loc 1 (state), loc 2 (state), … loc M 2 (state) URL N Loc 1 (state), loc 2 (state), … loc M N (state) … … Routing Table ZGP0024 Content Distribution Networks