Final Overview EECS 122

Transcription

1 The Network ore Final Overview EES epartment of Electrical Engineering and omputer Sciences University of alifornia erkeley any interconnected subs any different architectures dvertises a service to the end devices E.g. Phone v/s the Internet January, EES Lecture (KP) Review: heck List The internet consists of many s ig Picture Layers Where protocols are implemented Switching Techniques pplications NS HTTP STP Network Layer: Routing lass-ased; lassless ddressing ijkstra; ellman-ford GP Inside Router rchitecture: Input, Output Scheduling: Fairness, GPS, WFQ istributed lgorithms Overlay Networks any Internet Service Providers at each level of the Hierarchy Tier Tier Tier- Tier- Tier Tier- NP Tier Tier- Tier- January, EES Lecture (KP) January, EES Lecture (KP) The Network Edge: Example: ackbone Network end systems (hosts): run application programs e.g. Web, at edge of client/server model client host requests, receives service from always-on server e.g. Web browser/server; client/server peer-peer model: minimal (or no) use of dedicated servers e.g. Gnutella, KaZa, Skype January, EES Lecture (KP) January, EES Lecture (KP)

2 etropolitan rea Network Two fundamentally different ways to forward information ircuit Switched Information is exchanged in units of calls Network resources are reserved for the duration of the call Example: The Phone Network Once a call goes through, subsequent calls cannot degrade call quality Packet Switched Information is exchanged in units of packets Typically, no resources are reserved atagram: Each packet is forward independently Example: The Internet Virtual ircuit: ll the packets from a given stream take the same path through the Example: T, ISN, Intserv January, EES Lecture (KP) January, EES Lecture (KP) ampus Network Internet Layering pplication TP UP IP Network GP HTTP RTP TFTP TP UP IP lmost ny kind of application can write directly on IP Including new transport protocols IP cannot be avoided s long as the routers speak IP, any application that can make do with datagram service can be written and implemented on the end devices. No co-ordination, standards activity etc. is required!! Ethernet FI Token Etc. January, EES Lecture (KP) 8 January, EES Lecture (KP) Local rea Network message segment H t datagram H n H t frame H l H n H t source application transport link physical Encapsulation & Layers H l H n H t link H l H n H t physical switch H t H t H n H t H n H l destination application transport link physical H t H n H t H n H l link physical H t H n H t H n H l router January, EES Lecture (KP) 9 January, EES Lecture (KP)

3 pplication Protocols HTTP The ore provides a service to the hosts Host Host-Host: HTTP, STP Host-Network: NS Network-Network: Routing Protocols (e.g. OSPF) Host HTTP: hypertext transfer protocol Web s application layer protocol client/server model client: browser that requests, receives, displays Web objects server: Web server sends objects in response to requests HTTP.: Non Persistent HTTP.: Persistent P running Explorer ac running Navigator HTTP request HTTP response HTTP request HTTP response Server running pache Web server January, EES Lecture (KP) January, EES Lecture (KP) NS Features berkeley eecs argus sims root edu com gov mil org net uk fr mit Hierarchical Namespace istributed architecture for storing names Nameservers assigned zones of the hierarchical namespace ackup servers available for redundancy dministration divided along the same hierarchy NS client is simple: Resolver lient server interaction on UP Port (but can use TP if desired) No pipelining: Persistent HTTP server leaves connection open after sending response TP overhead minimized subsequent HTTP messages between same client/server sent over open connection client issues new request only when previous response has been received one RTT for each referenced object Pipelining: client sends requests as soon as it encounters a referenced object as little as one RTT for all the referenced objects default in HTTP/. January, EES Lecture (KP) January, EES Lecture (KP) How does a name get resolved Query walks its way up and down the hierarchy Iterated query I don t know, but here s who to ask next Recursive query I don t know right now, but I ll get back to you Network Layer ontrol Functions: Ensure that routers are configured to deliver packets correctly to the destination Path Selection (called routing in the book) onnection Setup: required in virtual circuit routing. ata Functions: Ensure that arriving packets are forwarded correctly within a router with minimum delay Forwarding January, EES Lecture (KP) January, EES Lecture (KP) 8

4 Interplay between path selection and forwarding IR: Example value in arriving packet s header path sel. algorithm forwarding table header value output link path selection algorithms run as application protocols forwarding is a function mostly Implemented hardware Routing = Fwding + Path Selection Example 8../ ommon prefix is bits: Number of addresses: 9 = Prefix aggregation ombine two address ranges 8../ and 8../: gives 8../ Routers match to longest prefix January, EES Lecture (KP) 9 January, EES Lecture (KP) lass-base ddressing ddressing reflects internet hierarchy bits divided into parts: lass lass lass 8 host host host ~ million nets hosts Link State Protocols. Every node learns the topology of the Flooding of Link State Packets (LSP). n efficient shortest path algorithm computes routes to every other node. Node updates Forwarding Table January, EES Lecture (KP) January, EES Lecture (KP) IP addressing: IR Route omputation: ijkstra IR: lassless Interomain Routing net portion of address of arbitrary length: subnet address format: a.b.c.d/x, where x is # bits in subnet portion of address subnet host part part.../ Every node knows the graph ll link weights are >= Goal at node : Find the shortest paths from to all the other nodes. Each node computes the same shortest paths so they all agree on the routes Strategy at node : Find the shortest paths in order of increasing path length List the nodes in increasing order of (shortest) distance S(k): closest k nodes Iteration k yields S(k) and a way to get there S()={} S()={,} S()={,,} January, EES Lecture (KP) January, EES Lecture (KP)

5 istance Vector lgorithms Nodes communicate distance estimates to their neighbors, not topology information ased on the ellman Ford Equation: efine (x,y) to the shortest distance from x to y. (x,y) = min vεn(x) {c(x,v) + (v,y)} where N(x) are the neighbors of node x. Why is this true? Let (x,v,y) be the shortest path from x to y where the first node after x is v. Then (x,v,y) = c(x,v) + (v,y). (x,y) = min v (x,v,y) = min v {c(x,v) + (v,y) } Router rchitecture Overview Two key router functions: run routing algorithms/protocol (RIP, OSPF, GP) forwarding datagrams from incoming to outgoing link Interconnection fabric January, EES Lecture (KP) January, EES Lecture (KP) 8 istance Vector: link cost changes The Forwarding ecision Process Link cost changes: node detects link cost change updates routing info, recalculates distance vector Good news travels fast but ad news can travel very slowly.ounting to infinity x y z atagram Routing: Each packet is independently forwarded at each router ust look up IP address ranges atch Longest Prefix Virtual ircuit Routing: call setup, teardown for each call before data can flow each packet carries V identifier (not destination host address) every router on source-dest path maintains state for each passing connection link, router resources (bandwidth, buffers) may be allocated to V (dedicated resources = predictable service) January, EES Lecture (KP) January, EES Lecture (KP) 9 GP Pairs of routers (GP peers) exchange routing info over semi-permanent TP connections: GP sessions GP sessions need not correspond to physical links. When S advertises a prefix to S, S is promising it will forward any datagrams destined to that prefix towards the prefix. S can aggregate prefixes in its advertisement Output Queued Routers c a b S a S c a b c S b d egp session igp session January, EES Lecture (KP) January, EES Lecture (KP)

6 Input Queues: Head-of-line locking The packet at the head of an input queue cannot be transferred, thus blocking the following packets Input Input Input annot be transferred because of HOL blocking Output Output Output annot be transferred because of output contention elays and Queues rrival Rate: α = (t)/t, t vg Packet elay: t = (()+()+.+(n))/n as n Packets Q(t): # Packets in the system at t Sender vg Occupancy = rea Q(s) Receiver (t): at point at point vg # packets in sys: Q = (Shaded area)/t as T Wastes router capacity s T time January, EES Lecture (KP) January, EES Lecture (KP) omponents of Per Hop elay Little s Law Propagation delay: time it takes the signal to travel Only random component from source to destination Packet transmission time: time it takes the sender to transmit all bits of the packet Queuing delay: time the packet need to wait before being transmitted because the queue was not empty when it arrived Processing Time: time it takes a router/switch to process the packet header, manage memory, etc Shaded rea up to time T is equal to both. ()+()+.+((T)). T Q(t) dt ivide and multiple by (T):. [()+()+.+((T)) / (T)] (T) ivide both (rewritten) and by T and take limits Q = α average occupancy = (average elay) X (average arrival rate) January, EES Lecture (KP) January, EES Lecture (KP) elays and Queues Flows and Fairness ax-in Fair llocation Packets Sender Receiver (t): at point at point b/s b/s R Want to treat all the flows as equally as possible.. b/s time. b/s Give the full.b/s and get.b/s each (.,.,.) January, EES Lecture (KP) January, EES Lecture (KP)

7 echanisms to Improve est Effort lassification and Scheduling rop Policies all admission Policing Implementing even a subset of these can help! Performance Guarantees: Flows+Policing+Scheduling Policing Scheduling arriving traffic token rate, r bucket size, b per-flow rate, R WFQ = b/r max January, EES Lecture (KP) January, EES Lecture (KP) dvanced Queuing Functions Packet classification: map each packet to a predefined class use to implement more sophisticated services (e.g., QoS) Flow: a subset of packets between any two endpoints in the lassifier class class class n uffer management Scheduler FFS Priority Round Robin WFQ January, EES Lecture (KP) 8 odeling Issues Error correction ssume that errors can eventually corrected Propagation elay Fixed Variable but no more than d Variable with no upper bound Other components of delay Queueing elay Transmission elay Packet order FIFO an be delivered in arbitrary order January, EES Lecture (KP) Policing echanisms aintaining accurate topology information Token ucket: limit input to specified urst Size and verage Rate. its end points send messages to all their neighbors who then flood. slow link bucket can hold b tokens tokens generated at rate r token/sec unless bucket full over interval of length t: number of packets admitted less than or equal to (r t + b). January, EES Lecture (KP) 9 January, EES Lecture (KP)

8 aintaining accurate topology information aintaining accurate topology information own own its end points send messages to all their neighbors who then flood. fails own its end points send messages to all their neighbors who then flood.. fails marks the link down. comes back up January, EES Lecture (KP) January, EES Lecture (KP) aintaining accurate topology information aintaining accurate topology information own own its end points send messages to all their neighbors who then flood.. fails marks the link down own its end points send messages to all their neighbors who then flood.. fails marks the link down. comes back up marks the link up. marks the link down January, EES Lecture (KP) January, EES Lecture (KP) aintaining accurate topology information aintaining accurate topology information own its end points send messages to all their neighbors who then flood.. fails marks the link down. comes back up own msg lost This can be fixed with sequence numbers, but then other problems emerge its end points send messages to all their neighbors who then flood.. fails marks the link down. comes back up marks the link up. marks the link down. fails message lost thinks is down when it is actually up! January, EES Lecture (KP) January, EES Lecture (KP) 8 8

9 Synchronous v/s synchronous lgorithms Synchronous algorithms can be described in terms of global iterations. The time taken for a given iteration is the time taken for the slowest processor to complete that iteration: time driven E.g. T or SONET synchronous algorithms execute at a processor based on received messages and internal state: event driven E.g. IP protocols which must run over heterogeneous systems ompare with Synchronous Slot size is affected by the slow node Node Node Node January, EES Lecture (KP) 9 January, EES Lecture (KP) Implementing a Synchronous lgorithm Suppose the slowest process can complete an iteration in time T p Link delay is always less than T l Then a slot size of T p +T l or more is sufficient ut most processors may be most of the time What if T p and or T l are not known? synchronous computation No notion of slot size at all! 8 9 Node Node Node Why should this work? January, EES Lecture (KP) January, EES Lecture (KP) Local Synchronization Send update k after you ve heard update k- from all neighbors. Node Node Node Why bother with synchronous lgorithms To reduce the synchronization penalty ifficult to get the synchronous algorithm to start The is dynamic Flows Topology Think of the algorithm having to restart with a new set of initial conditions, every time there is a failure hanges create events which may or may not have global impact Event-driven algorithms better suited January, EES Lecture (KP) January, EES Lecture (KP) 9

10 Soft State ontent ddressable Network (N) State with Time-Out Example: host joins a group by sending a join message to a host manager. The manager adds the host to the group for the next T seconds. If the host wants to stay in the group it must send a refresh message within T seconds to the manager. Otherwise it is dropped. dvantage: anager robust to host failure isadvantage: Too many messages ost internet protocols use this way of communicating Trades of simplicity of correctness with complexity of communication ssociate with each node and item, a unique id in an d-dimensional space Example for d=: node might be called (,,) Example for d=: song might be called (,,) Properties Routing table size O(d) Guarantee that a file is found in at most d*n /d steps, where n is the total number of nodes January, EES Lecture (KP) January, EES Lecture (KP) 8 Kinds of Overlay Networks N Example: d= Two kinds of Overlays. Only Hosts: Peer to Peer Networks (PP) Example: Napster, Gnutella, KaZa. Only Gateway nodes: Infrastructure Overlays ontent istribution Networks (Ns) Example: kamai Overlay node structure Regular: N dhoc: Gnutella Hybrid: KaZa Functions Route Enhancement: etter QoS, pplication Level ulticast Resource iscovery: PP Space divided between nodes ll nodes collectively cover the entire space Each node covers either a square or a rectangular area of ratios : or : Example: ssume space size (8 x 8) Node n:(, ) first node that joins cover the entire space n January, EES Lecture (KP) January, EES Lecture (KP) 9 ontent ddressable PP Networks (N) N is one of several recent PP architectures that imposes a structure on the virtual topology uses a distributed hash-table data structure abstraction Note: item can be anything: a data object, document, file, pointer to a file routes queries through the structured overlay attempts to distribute (object, location) pairs uniformly throughout the supports object lookup, insertion and deletion of objects efficiently. Others: hord, Pastry, Tapestry overed space divided between nodes Node n:(, ) joins space is divided between n and n n n January, EES Lecture (KP) January, EES Lecture (KP)

11 Nodes continue to join Node n:(, ) joins space is divided between n and n Node n: (,) n N Example: Two imensional Space Each item is stored by the node who owns its mapping in the space n n n f n n n f f n f January, EES Lecture (KP) January, EES Lecture (KP) Nodes continue to join N: Query Example Nodes n:(, ) and n:(,) join n n n n n Each node knows its neighbors in the d-space lso knows the d-space controlled by its neighbors Forward query to the neighbor that is closest to the query id Example: assume n queries f n f f n n n n f f January, EES Lecture (KP) January, EES Lecture (KP) Items are also mapped in the same space Items: f:(,); f:(,); f:(,); f:(,); n f f n n n f n f Infrastructure Overlays Overlay users are not directly connected to the overlay nodes E.g. kamai January, EES Lecture (KP) January, EES Lecture (KP)

12 Overlay Routing: Edge apping Overlay oncept: Going own R? IP() Overlay users are not directly connected to the overlay nodes E.g. kamai User must be redirected to a close by overlay node Edge-apping, or redirection function is hard since # potential users enormous User clients not under direct control 8 Need this link to be very reliable and fast! January, EES Lecture (KP) January, EES Lecture (KP) Overlay Routing: Edge apping IP Network is the Overlay? IP() Overlay nodes interconnect clients Enhance nature of connection ulticast Secure Low Loss uch easier to add functionality than to integrate into a router a c 8 d b IP Routers and attach to a virtual circuit e.g. T The IP sees the virtual circuit as a link This is called Link Virtualization and is commonly deployed January, EES Lecture (KP) 8 January, EES Lecture (KP) Overlay Routing: dding Function to the route Overlay nodes interconnect clients Enhance nature of connection ulticast Secure Low Loss uch easier to add functionality than to integrate into a router Overlay nodes can become bottlenecks January, EES Lecture (KP) 9