Computer Networks 2002/2003

1. Computer Networks2002/2003 Introduction Johan Lukkien Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

2. Overview • Motivation, background • Terminology • Physical organization • topology • broadcast networks • limitations & scaling Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

3. Motivation • Connect independent systems • “Loose” connection • communicating partners also operate stand-alone • Openness • explicit boundaries: interfaces, protocols • separate development • Scalability • size, location (spatial), administration • Share resources • connect resources & users • share the infrastructure; separated from applications Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

4. What is a network? • Physically: physical infrastructure, connecting access points • infrastructure: cables, repeaters and hubs • access point: physical location where a terminal (a user of the network) attaches • Logically: a facility for the exchange of information between disparate applications • disparate: unrelated, not sharing memory Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

5. Hardware concepts • A network is modeled by a graph • Nodes: Processors, Memories, P+M • P+M may also be a dedicated network component • Connect P with M: tightly coupled • Connect P+M’s: loosely coupled • Edges represent communication medium • broadcast, i.e. multiple access • e.g. bus, ethernet • point-to-point • e.g. “switched’ ethernet Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

6. Broadcast medium • All access points see the same physical signals • ...though not at the same time • introduces dependence on propagation delays • need to deal with “collisions” (in fact: interference) • arbitration: centralized / distributed • try again.... • Implementation: • Several access points on a single “wire” • e.g., bus, coax ethernet, wireless LAN • Access points connected by repeaters & hubs • hub: repeat and amplify signal (as well as noise) on all lines, except where it came from • “layer 1 switch”, e.g. for ethernet • repeater: two-party hub (copy signal from one onto other) • The above two are functionally identical Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

7. Bus-based • Arbitration: master/slave • Tightly coupled Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

8. 10Base2 ethernet • No arbitration: detect and avoid collision CSMA/CD • 10Base2: 10 Mbps, 200meter, coax • Loosely coupled Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

9. Network topology issues • Maximal distance between any two nodes: diameter • Minimum amount of traffic/wires that can go between any division in two halves: bisection bandwidth • Number of neighbors of a node: degree • Inter-dependence of degree, number of nodes and diameter • maximize #nodes for given degree and diameter • Moore upperbound • roughly logarithmic relation nodes and diameter Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

10. Examples: grid and hypercube • (a): degree: 4, diameter: sqrt(N), BW: sqrt(N) • (b): degree: log(N), diameter: log(N), BW: N/2 Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

11. Examples: tree and star • Tree: degree: 3, diameter: log(N), BW: 1 • fat tree: scale bandwidth per level • Star: degree 1 or N-1, diameter: 2, BW: N/2 or 1 Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

12. Terminology • Link: a physical medium for communication • bi-directional, uni-directional • Circuit: sequence of links • Channel: a means of communication (physical or logical) • Path: sequence of channels • The throughput or bandwidth of a channel is the number of bits it can transfer per second • The latency or delay of a channel is the time that elapses between sending information and the earliest possible reception of it. • Full duplex communication: concurrently, both ways along the channel; vs. half duplex Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

13. Limitations of broadcast medium • Capacity shared by connected nodes • though average and peak demands may vary • ....hence, not too many nodes • Propagation delays • for collision detection • colliding parties need to be able to detect a collision during transmission; ...hence, propagation delay and speed of medium determine minimum packet size • arbitration: just long negotiation delays • .....hence, limit physical distances • Robustness: single point(s) of failure • network-wide physical effects of attach/detach/fail • ....hence, decouple Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

14. Alternatives • Establish a circuit only when needed: circuit switching • e.g. (older) telephony, “switched” networks • possibly, simultaneous connections • guarantees certain quality, makes billing easy • resource in-efficient, especially for short-lived interactions • vulnerable to errors and failures Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

15. Switching and sharing Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

16. Switching networks • For tightly coupled case: “backplane” for a multi-computer • Several concurrent connections possible (crossbar) Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

17. Alternatives • Establish a circuit only when needed: circuit switching • e.g. (older) telephony, “switched” networks • possibly, simultaneous connections • guarantees certain quality, makes billing easy • resource in-efficient, especially for short-lived interactions • vulnerable to errors and failures • Have physically separated networks and exchange packets rather than signals • admits more concurrency, in principle • in the limit: networks of size 2 (point-2-point link) • no interference anymore • no direct link between communicating partners! • must know what a packet is • can connect heterogeneous networks Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

18. Heterogeneous • Connect multi-computers, PC’s, palm-tops • using a variety of physical connections • Just connect arbitrary networks • e.g. wire-less and regular LAN • Generally, • physical connections do not connect all pairs directly • some nodes need to transport information explicitly: store & forward packets • these nodes need to interpret (and, hence, understand) the passed information to some extent Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

19. Conclusions • Networks based on broadcasting are limited in size (geographical, number of nodes) • Collision domain: for collision-based methods, the domain in which physical broadcasting is done • Communication based on packets for • scaling in size • decoupling Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking

20. Exercises • My laptop’s hard disk is 30 GB. I cycle home in 30 minutes. What are throughput and latency of this “channel”? • Exercises 3, 4, 6 on page 81 and 35 on page 83 • What are bisection bandwidth, degree and diameter of • a line network • a ring • a torus (the cartesian product of two rings; draw it first) • Why are not all computers of the world connected in a single broadcast domain? Mention several reasons. Johan J. Lukkien, j.j.lukkien@tue.nl TU/e Computer Science, System Architecture and Networking