Internet Architecture and Experimentation Today l Internet architecture l Principles l Experimentation
A packet switched network Modern comm. networks are packet switched Data broken into packets, packet routed separately from source to destination Simpler than circuit switching a physical circuit is reserved from src to dest (e.g., buffer space) Links and routers Links physically move packets from place to place Routers receive packets from incoming links and put them on outgoing ones toward the destination 2 2 2 Src 1 3 1 2 3 1 2 3 1 3 1 3 Dest 2
Networking protocols For communication to work, parties must agree on several things How does the receiver knows it got the last bit of a msg? How longs are integers? To handle complexity protocols organized into layers Protocols set of rules and formats to use Sequence of messages to exchange Format of messages Implemented by pair of software modules on each end Typically arranged in layers protocol stack 3
Protocols in communication Internet protocols organized into four layers ISO OSI, 7 layers, not used in practice but good for understanding As a message is Pass down, each layer adds a header (sometimes a trailer) On the receiver side the message is push upward with each layer stripping off and examining their own headers Source Destination 7.Application 4.Transport 3.Network 1.Physical Communication medium 4
Internet protocols Link layer Ethernet, WiFi, Move packets between locations Physically interface with communication medium Network layer - IP Move packets between hosts, over sequences of links Provided mostly by routers as they do packet forwarding Transport layer TCP, UDP Transfer msgs between clients, including breaking them into packets, controlling low, re-ordering, etc Application layer SMTP, HTTP, P2Ps, DNS, Implements each particular application (email, WWW, ) Application and transport layer protocols are end-toend - intermediate nodes do not participate 5
Internet protocols TCP/IP IP Internet Protocol primary network layer protocol A best effort service unreliable, no guarantees Somebody else does reliability TCP Connection bet/ a host and network is called interface Each has an address; format defined by IP; in IPv4 addresses are 32b expressed as four decimals Address are assigned in a manner that reflects network organization IP addresses often grouped by their prefixes initial set of bits e.g. 165.124.0.0/16 addresses with first 16b = 165.124 6
Internet protocols TCP/IP TCP to ensure data reliability Basic service a connection, communication channel between two hosts Protocol specifies how connections are initiated (three-way handshake), used and released General approach to reliability acknowledgements After sending each segment, source sets a timer waiting for an ack Congestion avoidance implemented by varying number of un-acks packets the sender allows (window size) 7
Design principles of the Internet To meet functional goals of the Internet, several design decisions have been made inferred principles Decentralized design and operation The Internet is a loose interconnection of networks, not really one network A key factor for its rapid growth IP hourglass or IP over everything Overarching goal to provide connectivity key for this is IP Many protocols at the application or transport layer, many communication technologies underneath email www phone SMTP HTTP RTP TCP UDP IP Ehternet PPP CSMA async sonet Copper radio fiber 8
Design principles of the Internet Stateless switching Switches are expected to be stateless wrt connections When packet arrives, router inspect packet IP s header and a table to determine how to forward the packet Results in very simple routers, but complicates measurements Simplicity of router element is related to End-to-end argument Insight many network functions require cooperation form end-systems for correct and complete operation Don t try to provide it within the network (expect for performance reasons), since this will not simplify end-systems E.g. reliable transfer will always need to be checked by end-systems Some challenges to end-to-end: untrustworthy world, more demanding apps (use of CDNs), less sophisticated users, 9
Details of Internet operation Basic components of the Internet end-systems/ hosts, links and routers End-systems the client/servers in a client-server model Routers intermediaries that direct packet traffic through the network; connected to multiple links Links the media that physically move packets bet/ places Autonomous systems (ASes) Internet is a collection of separately, usually competing, managed networks Individual network elements are organized into AS set of network elements under a single organization s control ASes exchanged traffic at connection points peering or exchanged points Connections formed by establishing a link between routers in each AS, called gateway routers 10
Details of Internet operation routing Packets must be forward to its destination as specified in each packet s header Routing is a graph theory problem find the lowest-cost path between two nodes Routers decide outgoing interface for each packet Decision is called routing Placing it in the correct outgoing interface forwarding Set of links used packet s path For routing, routers keep info on network config and state Update state of links (added, removed, failed router, ) Internet routing is hierarchical Two levels intra-as (intra-domain, or interior gateway) and inter-as (inter-domain) Scalability and independence 11
Intra-domain routing Common intra-as protocols Distance vector, Bellman-Ford, algorithms RIP Link state algorithms OSPF, IS-IS Routing Information Protocol (RIP) Routers inform neighbors of topology changes periodically Router s routing table A record for each destination Fields: outgoing link, cost (e.g. hop count) A link that is down has an cost Routers send table summary periodically (30s) or when an update from other routers makes it change its tab As network grew in size, there was a burst of traffic every 30 despite routers being initialized at random times 12
Intra-domain routing Link-state Routing (OSPF) Router talks to everyone, but only tells them what it knows for sure (state of directly connected links) Distance vector router talks only to its neighbors, but tells them everything it has learned Routers maintain a network map Updated when links change Each router monitor the links to which is connected and initiates a flooding protocol upon change Message exchange is called Link State Advertisement To compute path to destination Each computes best path from it to every destination path with the lowest sum of link weights è routing table OSPF Open shortest path first (Dijkstra s) 13
Intra-domain routing Inter-domain routing is a bit more challenging Economic and policy issues e.g. I prefer to send traffic via AS X than Y, but I ll use Y if it is the only path, and don t want to carry traffic between X and Y Border Gateway Protocol (in v4) some highlights Internet is an arbitrary interconnected set of ASes BGP advertises complete paths as an enumerated AS list Necessary to enable policy decision and detect loops Designed to run on top of TCP So no need to retransmit, but need keep-alive messages BGP updates A BGP router will advertise (no obligation) one of several routes to a destination, selected according to its policies It can also send a negative advertisement withdrawn route 14
Measurement and experimentation At every layer Network infrastructure and routing Traffic Applications And higher-layer concerns Censorship Ethical concerns Where can take measurements? Application Transport Network Link 15
Measurement locations in an ISP ISP X Gateway routers route traffic between ISP and other AS Access routers provide connectivity to customer networks over an access link Gateway router Access router NAP (Network access point) Peering links Backbone router Access link An exchange point of multiple ISPs that can exchange traffic ISP Y Gateway, backbone and access routers communicate with each other using an intra-as protocol Customer 16
Measurement in the wide area Heterogeneity and reproducibility Measurement at a single location are hard to generalize Many high-res internal measurements are kept internal Wide area measurement locations Many of the same places, but across a wider area 17
Internet model National Backbone Operators Sprint, MCI, AGIS, NAP NAP Regional Access Providers Local Access Providers ISP1 ISP2 Customer IP Networks 18
Updated Internet model Global Internet Core Global Transit/ National Backbones Hyper Giants Large Content, Consumer, Hosting CDN IXP IXP IXP Regional / Tier 2 Providers ISP1 ISP2 Customer IP Networks Labovitz et al., SIGCOMM 2010 Flatter and much more densely interconnected Internet Disintermediation between content and eyeball networks New commercial models between content, consumer and transit 19
Measurement in the wide area Some obvious issues Modeling the network, placing landmarks, designing multi-site measurement platforms, coordinating measurements, ensuring representativeness Some measurement platforms E.g., NIMI, Planetlab, Dimes, Archipelago, Dasu Representativeness Networks, applications, clients Observation and control Observation studies Controlled experiments Natural experiments 20
Coming up Experimental platforms and experimental design PlanetLab and SatelliteLab, Dasu Experimental designs Observational studies Conviva Controlled experiments Internet optometry, web search personalization Natural experiments Broadband markets and user behaviors Context Usage behavior and the importance of context 21