Advanced Computer Networks Spring Set #1

Transcription

1 Advanced Computer Networks Spring Set #1 Prof. Zygmunt J. Haas Computer Science Department The University of Texas at Dallas ECSS Richardson, TX Copyright 2019 by Zygmunt J. Haas All right reserved. No part of these notes may be reproduced, in any form or by any means, without permission in writing from the author. Introduction and Motivation Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

2 The Course Objectives To understand the current state-of-the-art of architecture of communication network protocols, design, and applications To understand the current and future challenges in the design and implementation of communication protocols and communication networks The goal is to educate the next generation of network researchers, rather than network programmers Wireless Networks Laboratory Copyright by Zygmunt J. Haas, The Objectives if Communication Networks To support communication between and among applications on different devices (e.g., computers, PDA, cell phones, etc) To do so, one needs to understand the application needs and requirements, such as: Data rates Traffic patterns (e.g., bursty vs. constant bit rate) Traffic destinations (e.g., unicast, multicast, broadcast, anycast) Quality of Service (QoS) parameters (e.g., delay, jitter, loss) A network is a collection of nodes interconnected by links: Communication between adjacent nodes (i.e., across links, wired, wireless) Creating a network (i.e., interconnecting links, internetworking) Routing across the network Matching applications requirements Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

3 Topology Control (con t) Advantages of distributed vs centralized topology: Improved reliability Better availability Possibly better performance (lower delay, higher throughput, etc) Better growability patterns Advantage of centralized vs distributed topology: Reduced wiring Better manageability Simpler repair and maintenance Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Communication Link as a Pipe Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

4 A Link as a Communication Pipe A communication link/path is modeled as a conduit carrying liquid (water). C [bits/sec] T [bps] s [m/sec] T [bps] L [m] prop [sec] B [bits] Wireless Networks Laboratory Copyright by Zygmunt J. Haas, The Communication Pipe (con t) The bit capacity of the pipe, B [bits], the propagation delay of the pipe, prop, and the pipe utlization,, are given by: L T prop ; B C ; s C Thus a total delay of a message of size M [bits] is: total prop transmission For example, consider a link of 2/3 the speed of light: prop [sec] 41.7[ msec] Wireless Networks Laboratory Copyright by Zygmunt J. Haas, L s M C. 4

5 The Communication Pipe (con t) Now consider the case of a message of 10 [Kbytes] and link capacity, C=10[Kbps]: B 10[ Kbps] 41.7[ msec] 417[ b]. link capacity, C=100[Mbps]: transmission transmission [ b] 8[sec]; [ bps] [ b] 0.8[ msec]; [ bps] B 100[ Mbps] 41.7[ msec] 4.17[ Mb]. Wireless Networks Laboratory Copyright by Zygmunt J. Haas, The Communication Pipe (con t) [Kbps] pipe 1 [Gbps] pipe. bursty traffic Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

6 Satellite Communications Geo-stationary satellite (a reflector in the sky): 36,000[km] 250 [msec] one-way propagation time If the pipe is thin and short, thickening the pipe will reduce total delay, but if the pipe is long, thickening the pipe won t help! Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Matching the Pipes Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

7 Variations of voice traffic demand Office traffic variations during a typical weekday 1.0 Business office Originating calls (x10,000) Residence office Time of the day 12:00am 4:00am 8:00am 12:noon 4:00pm 8:00pm 12:00am Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Voice vs. Data CBR = Constant Bit Rate (e.g., uncompressed voice) VBR = Variable Bit Rate (e.g., data, compressed video) Traffic arrival VBR traffic CBR traffic Time Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

8 Voice vs. Data Why do we need to worry about data communications? Why aren t voice-based systems sufficient? Voice Data Real time (low delay) Delays are acceptable Sensitive to jitter Insensitive to jitter Some errors allowable Errorless communication Constant bit rate Variable bit rate burstiness Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Voice vs. Data (con t) Circuit switching was traditionally used to switch voice communication. Circuit switching involves a set up procedure, during which a resource dedication is performed. Once a circuit is establish, data can flow freely without any delay. Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

9 Circuit Switching: Switching Techniques Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Source 0 Time-Space Diagram Switch 1 Switch 2 Destination space propagation delay transmission time queuing+processing delay time Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

10 Source 0 Circuit Switching (set-up phase) Switch 1 Switch 2 Destination space propagation delay set-up message transmission time queuing delay Set-up delay time... Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Source 0 Circuit Switching (transmission & termination phases) Switch 1 Switch 2 link propagation delay Destination space transmission time termination delay time Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

11 Voice vs. Data (con t) But circuit switching assumes constant traffic patterns. Data communications, and especially computer communication, is very bursty. Such burstiness leads to inefficient use of network resources (e.g., a line is reserved, but is unused for the duration of long periods). The pipe remains empty most of the time!!! This is where switching (or message switching) was invented. Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Voice vs. Data (con t) The idea behind switching is simple - use statistical (time) multiplexing of the resources. The main problem: as the demand for resources can be highly variable in time (i.e., large degree of burstiness) and unpredictable, some data may not find the resources needed. An idea: since delay is not critical for data buffer the excess demand. But buffering may introduces unfairness in resource usage and large delay variations (jitter).furthermore, some traffic may be lost due to congestion. Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

12 Packet Switching The switching has been born! Segment the data into small units, s. Data to be transmitted H H H H H H a network H H Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Packet Switching: Switching Techniques Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

13 Switching Techniques Packet Switching (Statistical Multiplexing) Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Source 0 Message Switching Switch 1 Switch 2 propagation delay Destination space message transmission time total delay queuing delay time Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

14 Source 0 Packet Switching Switch 1 Switch 2 propagation delay transmission time queuing delay Destination space total delay time Wireless Networks Laboratory Copyright by Zygmunt J. Haas, Statistical Multiplexing Gain Assume a 1[Mbps] link and individual flows of 0.1[Mbps], active 10% of the time. Thus, when a flow transmits alone, it uses 10% of the link capacity. With circuit switching, we can accommodate 10 users, and the average utilization of the link will be 10%. With statistical multiplexing, we can support ~30 users with probability of < 0.1% (0.001) that more than 10 flows are active at the same time. With statistical multiplexing, we can better fill the pipe. Why cannot we fill the pipe completely (i.e., get 100% utilization, =1)? Wireless Networks Laboratory Copyright by Zygmunt J. Haas,

15 Statistical Multiplexing Gain Conclusion, with statistical multiplexing, we can accommodate significantly large number of users with small degradation in performance what is the performance degradation? But, there is no free lunch. Wireless Networks Laboratory Copyright by Zygmunt J. Haas,