Special Topics ECE 8833
Popular in Course
Popular in ELECTRICAL AND COMPUTER ENGINEERING
This 0 page Class Notes was uploaded by Cassidy Effertz on Monday November 2, 2015. The Class Notes belongs to ECE 8833 at Georgia Institute of Technology - Main Campus taught by Staff in Fall. Since its upload, it has received 7 views. For similar materials see /class/233909/ece-8833-georgia-institute-of-technology-main-campus in ELECTRICAL AND COMPUTER ENGINEERING at Georgia Institute of Technology - Main Campus.
Reviews for Special Topics
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 11/02/15
ECE 8833 Feb 26 2003 VOIP Reading RTP RFC s 1889 1890 SIP 3261 3267 RTP RealTime Protocol IPUDP extension IPUDPRTP RTP is for general realtirne streaming of voice Video etc Streamed data Continuous possibly variable rate data representing some sort of timebased signal RTP format 1 2 3 O l 2 3 4 5 6 7 8 9 O l 2 3 4 5 6 7 8 9 O l 2 3 4 5 6 7 8 9 O l V2PX CC M PT sequence number timestamp synchronization source SSRC identifier contributing source CSRC identifiers Header Key Fields Payload Type May change during session ie switch audio encoding Set of standardized payload types Sequence Number Incremented by packet not byte Timestamp 32 bits time associated with rst byte of packet s data Format not speci ed ie resolution not standardized May be the same in multiple packets ie Video frame encoded across multiple packets SSRC amp CSRC for synchronizing with RTCP NOT for additional multiplexing 0 Use UDP Port numbers to separate streams to same source Fields cont Sequence Number Used to identify lost or recrdered packets Timestamp Used to calculate Jitter Possibly delay RealTime Control Protocol RTCP Used to distributed information about participants amp quality Not a full participant management protocol Distribute reception reports etc Network time H323 H323 is an ITU standard Based on H320 Video over telephone lines Supports Video audio phone other Plethora of underlying standards Video Coding H261 H263 Audio G711 G728 G7XX call transport call setup extensions Application sharing data stream T 120 H323 Functional Components H323 Terminal Aka client device endstation H323 Gatekeeper Directory Server CAC Logging Billing Call forwardingredirecting Coordinate amoungst multiple Gatekeepers H323 Gateway Translate to H320 or classic telephony H323 MCU Conference call unit H323 is multicast indifferent ECE 8833 March 24 2003 Private Addressing Assignment no NSZ projects Pick an IETF working group Not one that we have reviewed in detail Send me email if there is any doubt Review at least 3 major RFC s or Drafts Should represent central effort of group Prepare 40 minute powerpoint presentation To be given at end of semester Private Addressing 10000 8 16 1921680024 These addresses are NOT assigned to any public Internet host No Internet router should forward packets to these addresses 3 They may be used by private networks ie not connected to the Internet Address Overlap Multiple Independent Private networks may use the same private IP address for a host Since they re not connected there is no chance of con ict or ambiguity Size Private Network may be as small as two nodes one node should just use the loopback address Or can be an lntemational network Connected by for example leased lines 100008 is 16 million addresses Private Network with public server Inside private Outside Public Server Srvr is multihomed ie two network interfaces one connected and addressed for the private network one connected and addressed for the public Internet NOTE Server does not forward packets Common application company web or mail server Address Translation Network Address Translation NAT Map between two different addresses Often from a private address to a public address Two main types of NAT Host based address translation Port based address translation NAT types Host Based Onetoone mapping between host and addresses Mapping may change dynamic or be static 0 Dynamic Typically when mapping is idle for N minutes 30 minutes 0 Some similarity with DHCP Port based Onetomany mapping between address and host 1e one public address is used for many hosts Port numbers used to distingush Not used in static Host based NAT Generalized bidirectional NAT This allows for inside amp outside to use the same addresses Ie Inside is private network 1 using 1921680024 Outside is private network 2 also using 1921680024 Use 1012024 to represent other network General NAT con guration Inside private Outside Public Typical simpli ed NAT con guration Inside private Outside Public ECE 8833 TCP Congestion Domain Names DNS Jan 13th Some Reading RF CS 072 web page 1P UDP T CP Congestion DNS Measuring RTT Delay Can measure RTT Send a packet Wait for acknowledgement Note time it took 0 Don t measure packets that got sent twice Kam s Algorithm 0 or didn t get acknowledged Then wait twice the measured RTT to resend Dynamic RTT running average RTTnew new estimate of RTT RTTold previous estimate of RTT RTTpckt measured RTT for a sendgtack RTTnew a RTTold la RTTpckt 0 Usually take use approximately 90 for a Round trip is 90 old time 10 current packet RTT estimate RTTnew aRTTold 1aRTTpkt Enhanced scheme Van Jacobson 1988 Uses enhanced measurement Accounts for average plus variance RTTpkt often measured at 500ms precision Round Trip Timeout RTO RTO Avg 4Deviation Initialized to Avg 0 Deviation 6 seconds Retransmit after RTO And again at 2XRTO And again at 4XRTO And again at 8XRTO And so on doubling multiplier each time Until receiving an ack or nally terminating Packets Loss amp Delay Gigabit Etheme T1 Fast Ethernet Trivia Question Can the Source Send a packet at 500kbps Gigabit ethernet 1000000kbps Tl1500kbps Fast ethernet 100000kbps Answer NO Source send a packet at 1000000kbps Dest receives packet at 1000kbps Packets can t be padded Delay between packets can result in AVERAGE of 500kbps for multiple packets TCP Windows Network temporarily stores data while in transit Simple formula complex implications Average bandwidth B Average Delay D Stored data S SDXB DelayBandwidth Product DelayBandwidth Bandwidth DataDelay TCP acknowledges data sent If you limit the amount of unacknowledged data 1e sender quits sending when it reaches a certain number bytes that have not been acknowledged Then bandwidth is reduced accordingly This is the basis of TCP congestion control TCP Congestion Control Sender adjusts transmission rate based on packet loss Recognize loss by lack of acknowledgement rate amount of unacknowledge data Add new threshold to sender window CWND congestion window Upper bound on data that has been sent but not acknowledged TCP bandwidth algorithm Start with slow sending rate small CWND Grow CWND for every ack note that as CWND exceeds packet size multiple acks create geometric CWND growth When a packet is lost slow down to 12 current rate and grow slowly linear growth Slow Start Slow Start Uses congestion window CWND Measured in bytes Incremented in segment size MSS CWND initialized to one MSS Grows by MSS for each packet that is received and acknowledged before RTO Exponential growth Attempting to approximate CWND Delay X Bandwidth Indication of Lost packet C 0N GEST I 0N Note that not every successfully sent packet is ack d If 4 packets sent ie CWND gt 4 M88 and they are all received then the receiver rnay acknowledge only the last packet Therefore packet is lost if 1 RTO timeout occurs this could be a lost ack also but is treated as lost packet 2 A duplicate ack previously sent packet is received gt ltCongestion Avoidanceilt When a lost packet is detected Set SSTHRESH 12 min CWND RCVRWin New variable SSTHRESH is Slow Start Threshold If lost packet is due to timeout set CWND MSS Growing CWND on successful packet ack 0 Previously add MSS to CWND for every successful packet If CWNDltSSTHRESH add MSS Slow Start If CWNDgtSSTHRESH add lCWND Congestion Avoidance Note that congestion avoidance is linear growth slow start is W growth Fast Retransmit amp Fast Recovery More modi cations proposed 1990 VJ duplicate acks Indicates a packet was lost but later packets arrived CWND gtgtMSS OR packets were reordered along the way Receiver must ack immediately for incorrect packet cannot delay If three or more duplicate acks received Mild congestion one packet was lost others owing Don t overreact 0 Set SSTHRESH 12 CWND Retransmit the missing packet 0 Set CWND SSTHRESH 3 segments Add one segment to CWND for any more duplicate acks 0 On next nonduplicate ack set CWND SSTHRESH from above Continue with normal CWNDgtSSTHRESH congestion avoidance Network Implications Fairness For T CP In Internet backbonez congestion occurs when a large number of hosts exceed the link capacity Verses finding modem speed in a single user perspective Each active sender dynamically seeks to achieve the same level of packet loss More loss gt slower throughput Less loss gt faster throughput Extensions amp Tricks 0 Problems with synchronization What if queue lls up and drops many packets at the same time 0 Causes all TCP connections to slow down at the same time 0 Network is suddenly underutilized 0 Solution try to throw away a few packets before queues are completely full In special cases Can slow one application down by throwing away its packets more often Same as giving one application more queue space or more priority Explicit Congestion Noti cation 3 httpwwwieyforgrfcU c3168lxt Instead of dropping a packet set a congestion bit in the packet header 1 Negotiate ECN capable senderreceiver Use bit pair in IP header NOT TCP header Set bit pair to 10 or 01 to indication ECN capable Leave at 00 for not ecn capable 2 Queues set congestion bit pair to 11 when consistently near full This is a noti cation to the receiver that the sendergt receiver link was full 3 Receiver sets new bit from RESERVED eld in TCP header receiver tells sender called ECNEcho ECE 4 Sender reacts to ECE bit in ack packet as if a packet was lost congestion avoidance without resending 5 Sender replies with TCP RESERVED eld CWR bit to let receiver know that it is responding to congestion ECE8833 Feb 24 2004 Differentiated Services DiffSerV Reading RFC 2474 Header bits RFC 2475 Architecture RFC 2597 Assured Forwarding RFC 2598 Expedited Forwarding QoS RSVP Maintains a large amount of state Router Tracks many ows RSVP sessions Router maps ows to queues Often one or two special queues per output Many ows per queue Router has to admit additional RSVP sessions based on capacity amp guarantees Trade off utilization vs level of guarantees New session may request different level of guarantee QoS local vs global RSVP hopbyhop Each router makes decision about CAC Router can at best only work on local optimization since it only has local knowledge of the network load Router has limited nonlocal topology information Via routing protocol but this does not re ect load QoS recap RSVP per node Track current allocation Admit new sessions based on loss amp delay speci cations Normal routing Map ow to appropriate queue Have to identify ow QoS Centralized control Note that routers are efficient at forwarding packets Centralize CAC Allocation Billing Allows perhaps better optimizations Pass minimum information to router LE What queue to use for each packet DiffServ Concept Setup info quot1 DiffSerV Adds Bandwidth Broker central control MarkingPolicing boundary function 0 Classi er Meter Marker Shaper Use Header bits Header bits can imply which queue to use Either directly Or simple mapping Reduces state information per router Functionality implemented at edge Traf c Differentiation Perhop Behavior PHB Identifiable overall QoS behavior Large number of potential PHB s PHB should be globally understood Code Point Differentiated Services Code Point DSCP Header bits used to obtain a PHB Code point is locally admin domain signi cant Note many unique PHB s are possible Finite number of code points Header Bits 0 Original 8 bit TOS eld implementing 3 bits for Type of Service Delay Throughput 111 defined 5 bits for precedence Normal critical Network not technically permitted to change TOS eld bits by early de nitions Renamed DS Field for DiffSerV IPV4 amp IPV6 Diff SerV Many attempts to de ne sets of PHBs Assured Forwarding Expedited Forwarding ECE 8833 UDP Congestion Domain Names DNS Classfull amp Classless addresses Jan 15th UDP Congestion UDP no feedback Sender doesn t know if packets reached source or was lost in transit Sending application frequently doesn t know if receiving destination amp program exists UDP does not slow down for congestion just high loss rate TCP amp UDP together TCP slows down to accommodate loss UDP doesn t UDP doesn t share well with TCP Domain Name System Mapping Host Names gt IP Addr Hierarchical System Boundaries are the dots in names Top level server knows how to nd servers for edu gov org com net etc Servers can be primary or secondary Host has to nd dns server by IP address DHCP integration Update names Addresses Network vs host IP address contains two parts Network Address Specifies which physical network port on a speci c router Host address Specifies which device on the physical network How do we identify the Network address vs the Host address Class based addresses Class A one octet given lXXX 127XXX 127 Class A ranges available some are reserved Each range is 224 addresseshosts 16 million Class B two octets given 1280XX thru 192255XX 214 ranges class B networks available 216 hosts per class B 65535 Class C three octets given l9300X thru 223255255X 221 ranges class C networks available 255 addresses hosts per range Class of network is implied by rst octet Subnets amp Netmasks Noone needs a Class A 16 million addresses Can we break a class A into smaller types Yes this is called subnetting Class A can be divided into class B size ranges or Class C size ranges Can also break a class B into 255 class C s Subnet has become generic term for locally connected hosts ie same Layer 2 topology Netmask 32bit mask which identi es the boundary between Network and Host portion Not carried as part of IP packet Allows the networkhost boundary to be identi ed in subnetted situation Netmask is ones for nework zeros for host All ones must be consecutive Slash notation Classfull addressing implies 8bit boundary CIDR Classless Internet Domain Routing Use arbitrary netmask instead of 8 bit boundary Backbone Internet Routers Backbone routers have a route entry for each network that is in use Something like 50000 route entries in current Internet Route entries have to be searchable in real tirne packet time Routing Basics Local subnet Compare Network portion of address to determine if an address is local or remote Host delivers directly to destination Typically using ARP to nd MAC address Remote subnet different network address Delivery through router IP host must know a router Typically called default router or gateway Routing Routers have two primary functions Forwarding packets L3 switching 0 Needs to be done in packet time many packets sec 0 15 5Mbs link 100000 packets per second each direction 0 HardwareASICFirmware Calculating routes where to forward 0 Needs to be done in network recon g time 0 Seconds 0 CPU Calculating RoutesgtRouting TablesgtForwarding Some Reading TCPIP Illustrated Volume 1 The Protocols W Richard Stevens 7 chapters 1 3 11 17 21 21 Jan 8th Packet Delivery Sockets I The Combination ofan address pair and port pair I Packets are not guaranteed to successfully arrive forms a socket at the IP destination 7 Source IP Address 7 Destination computer may be busy or of a Source Port number s Packets may get corrupted error in transmission 7 Destination IP Address 7 Network may temporarily run out ofcapacit Destina i m P0 mequot There are no traf cjams aicess trafflc lSjuSl thrown away I Originally referred to TCP but can be applied to loso UDP I Best effort Delivery Programming Concept keeps track of many 7 Packets can arrive out of order details once setup is complete 1P IP Packet Header Field 20 bytes options I IP header speci es IP address and basic packet options 20 byte header I UDP user datang protocol a 8byte header 7 Adds source port and destination port a Adds length and checksum error checking 7 Delivery still not guaranteed r Packets can arrive outoforder I UDP doesn t really make a connection UDP Header 8 bytes in on is on smn rm manoan rm loonuorungh norm uzkm om mun hugh TCP Transmission Control Protocol I Guaranteed delivery almost 7 will make repeated attempts to deliver 7 Failure will eventually result in an errornoti cation I 20 byte header 20bytes options 7 Source amp Destination Port 7 32bit sequence number 7 32bit sequence acknowledgement 7 window size ow control 7 Checksum I Requires a setup amp shutdownteardown I Makes a connection pointtopoint I TCP connections are bidjrectional TCP Header 20 bytes TCP Options TCP Sequence Number I Sequence number is key to delivery guarantee I Sequence number counts through the number of bytes successfully sent not the number of packets I Each Packet has as is sequence number the count of the rst byte in the packew data eld I Acknowledgement eld going the other way Acknowledges the highest number byte successfully received without gaps I Sequence number does not start with zero security Sender Window I Part of TCP stack Data ack d as received Data sent but not yet acknowledged Highest sequence number sent Space available for application data End of available space Receiver Window I Part of TCP stack Data used by application Datareceived but not yet used Highest sequence number received Space available for application data advertise this as window size End of available space Error Control Acknowlegements I Rule Receiver acknowledges the highest consecutive sequence number successfully received 7 Does not acknowledge gaps 7 Does not acknowledge packets w checksum errors 7 Does not send negative acknowlegement 39 NegatheAck ldon t have this data 7 Sender once a sequence number is acknowledged nothing less than that sequence number will everneed to be resent for the current connection 7 Note Acknowledgements can be lost Waiting for Acknowledement I How long to wait I Too long 7 We don t keep things moving quickly I Too Short 7 We might resend too soon I Would be nice to measure expected delay to acknowledgement Congestion I When queues get full packem get dropped this is congestion I If packet gets dropped due to congestion 7 Should not send more as this creates more congestion creates more drops 7 Should SLOW DOWN sending rate I TCP assumes lost packem are due to congestion 7 Slow down the sender When packets get lost Delay Fast link Slow link Sender Receiver not busy very busy How long to get from sender to receiver 7 Speed ofll ht Anorak zzc furphyslcal eable ApproximatelyB miemseenndsmile n s number orbits in queue Available Bandwidth bitsseeund ufuutbuund link is B Timetu send lastbitis Q B 7 Full lk llubytebuffaQ8lll kbsmudem 7 S nnn 7 Delay is 856 or W secund n 143 seemds 7 Queuing delays are umri nredum39nantll Queue Size varies 7gt 7 y vane Measuring RTT Delay I Can measure RTT 7 Send a packet 7 Wait for acknowledgement 7 Note time it took Don tmeasure packets that got sent twice Kam s Algorithm or didn t get acknowledged I Then Wait twice the measured RTT to resend I Dynamic RTT running average RTTnew new estimate ofRTT RTTold preVlOuS estimate ofRT RTTpcktmeasured RTT for a smd7gtack 7 RTl39neW a RTl39old la RTTpckt 7 Roundtrip is 90 old time 10 cunentpacket RTT estimate I RTTnew aRTTold l aRTTpkt I Enhanced scheme Van Jacobson 1988 I Uses enhanced measurement 7 Accounts for average plus variance I RTTpkt often measured at 500ms precision Round Trip Timeout RTO RTO Avg 4Deviation 7 Initialized to Avg 0 Deviation 6 seconds 0 Retransmit after RTO 7 And again at 2xRTO 7 And again at 4xRTO 7 And again at 8xRTO 7 And so on doubling multiplier each time 7 Until receiving an ack or nally terminating Slow Start 0 Recall Slow Start 7 Uses congestion Window CWND Measured in bytes Incremented in segment size MSS 7 CWND initialized to one MSS Grows by MSS for each packet that is received and acknowledged before RTO Exponential growth 7 Attempting to approximate CWND 7 Delay x Bandwidth Indication of Lost packet CON GES T I ON 0 Note that not every successfully sent packet is ac 7 If4 packets sent ie CWND F 4 MSS and ey are received then the receiver may acknowledge only the last packet Therefore packet is lost if 7 l RTO timeout occurs this could be a lost ack also but is treated as lost packet 7 2 A duplicate ack previously sent packet is received Congestion Avoidance 0 When a lost packet is detected N 7 If lost packet is due to timeout set CWND MSS GroWin CWND on successful packet ack Prevlously add MSS to cm or every successful packet 7 IfCWNDltSSTHRESH add MSS Slow Smn 7 IfCWNDgtSSTHRESH add lCW39ND Congestion Avoidance 7 Note that congestion avoidance is linear gowth slow s art is exponential growth Fast Retransmit amp Fast Recovery More modi cations proposed 1990 VJ duplicate acks 7 Indicates a packetwas lost but laterpackets arrived CW39ND gtgtMSS R theway 7 Recelvamust adquot immediately for mcorrectpacket cannot delay Ifthree or more duplicate ack received 7 Mild congestion one packet was lost others owing L et Set CW39ND SSTHRESH 3 segnents Add one segmenttu cwm fur any more duplicate aeks On neitnomdupllcate ack set CW39ND STHRESH from above 7 Conanue with normal CWNDgtSSTHRESH eongesaon avoidanee Network Implications Fairness In Internet backbone 7 congestion occurs When a large number of hosts exceed the link capacity Verses nding modern speed in asingle user perspective 7 Each active sender dynamically seeks to achieve the same level of packet loss 7 More loss gt slower throughput 7 Less loss gt faster throughput ECE 8833 Data Compression and Modeling Lecture 1 Introduction to Data Compression School of Electrical and Computer Engineering Georgia Institute of Technology Signal amp Coding Signal Continuoustime or discretetime function Scalar or vectorvalued Any informationbearing representations Coding due to Shannon Source coding conversion of signal into efficient digital representation for conservation of resources needed for transmission or storage of the signal Channel coding or error control coding transformation of signal or data so as to permit Spr39ng 2004 reliable communication in presence of noise or distortion spring2004 ECEEESSEHJWE CWWMM Lecturetti Slidetti spring2004 ECE39EEMEHMquot WWW Lecturetti Slidet Morse Code Alphabet A Framework for Data Compression bits A I Q Y e Transmissron B J R Z 7 Or Storage Source Encoder Media C K S 0 8 Channel D L T 1 9 E M U 2 Fullstop F N V 3 Comma G O W 4 Query H P X 5 Channel quota Comparison of source quot 3quot quot signaland reproduced v signalto determine Quality or Fidelity Spmgmm ECEEESSEHJuang CupyiighlZUUA Lecture 1 Slide 3 Spring 2004 ECEEEM E H quot9 WHEN Lecture 1 Slide 4 Data Com pression Various practical concepts related to time Time compression Time scale modification with or without changing the signal characteristics Garvey WD quotThe intelligibility of speeded speechquot Journal of Experimental Psychology 45102108 1953 Our interest Encoding or representation of information for storage ortransmission at the lowest cost in resources bandwidth storage area etc and without significant loss of information upon reconstruction Coding as a Task Representation of analog signal for digital transmission or storage often integrated with AlD conversion Compression of digital information to reduce transmission or storage requirement compression can also be realized in analog domain efficiency is defined in terms of bandwidth or storage required for the delivery of a xed amount of information such as a second of speech a video frame Result of coding is a sequence of digital often binary symbols The sequence of digital symbols may or may not have explicit delimiter Spring 2004 ECE39EE H H quot9 CWWH M Lecture 1 Slide 5 Spring 2004 ECE39EE H H quot9 WWW Lecture 1 Slide 0 From Shannon Information Theory lfthe minimum achievable source coding rate of a given source is strictly below the capacity of the channel then the source can be transmitted reliably by appropriate encoding decoding implicitly reliable transmission can be accomplished by separate source and channel coding lfthe source coding rate is strictly greater than the channel capacity then reliable transmission is impossible but we can still strive to reduce the negative impact of the rate excess by joint sourcechannel coding Memoryless block source codes can achieve minimum average distortion for a constrained rate in the absence of complexity constraint ie source coding subject to a fidelity criterion Issues in Source Coding Coding algorithm design Bit rate and distortion relationship lossy or lossless coding Implementation complexity Memory and delay requirement Robustness in performance against source variation Choice and significance of performance metric Impact of errors in code upon fidelity performance Practical coding algorithms often involve detailed tradeoffs among these issues Spring 2004 ECEEEM H H quot9 CWWH W Lecture 1 Slide 7 ECEVEESS a H Juang Cupyright 2004 Spring 2004 Lecture 1 Slide 8 Preliminaries Probability Theory Random Variables and Processes Linear systems Information Theory Entropy and measurement of information Shannon s SelfInformation LetXbe an event of a random experiment and PA denotes the probability that eventXwiII occur Selfinformation associated with eventXis given by iX10ngX IfXand Y are independent events PXYPXPY and thus iXY ilogb PXPY 710gb PX710gb PY iXiY When b2 the unit of information is called bit if the base is e the unit is nat if b10 the unit is hartley Sprrng 2004 ECE39EE H H quot9 CWWH M Lecture 1 Slide 9 Sprrng 2004 ECE39EE H H quot9 WWW Lecture 1 Slide 10 Information Source A source is an origin of information A random source is equivalent to a random experiment which generates outcomes for observation or reception The mechanism that a random source uses to generate information is usually unknown to the observer who sees only the outcomes of the experiment or the signals the source puts out As in random experiments an information source is associated with a probability measure from which one can calculate the entropy of the source When symbols or signals are generated in sequence the sequential experiments may or may not be independent Fundamental Dimensions of Source Coding Structure of information modeling How is information generated by the source How to approximate the informationgeneration process How to represent this process Random nature of information Efficiency of codes depends on how precise the knowledge the encoder has about the source How to estimate the source distribution How to design codes to achieve maximum efficiency given prescribed constraints Sprrng 2004 ECEEEM H H quot9 CWWH W Lecture 1 Slide rm Sprrng 2004 ECEEEM H H quot9 W EH W Lecture 1 Slide 12 Two Components of Information Defining a Source Parallel to Pr Space Structure deterministic component may or may not sample Spacei Observation Spacequot Signal Space be known may or may not be easily represented bu39it Upon a SymbOI set A aii1 Entropy random component never known which is also called an alphabet without loss of completely in real world generality the symbols a are referred to as letters and m the size of the alphabet XtAcoscot Vt Let XXX2X3X be a signal sequence generated by the source A sequence of length n so MaTny 39tncompifte waysfo Weir 39t t f generated can be considered as an outcome of a infgpeer g Ig dsgrzngfpzzmgn come 0 an combined experiment with the observation space Treat the amplitude of the sinusoid as random variable formed by the carteS39an prOdUCt 0f the original Treat the phase as random variable alphabet Aquot A X A gtltgtlt A and X a e A Treat the signal not as a sinusoid but a general random process Again the experiments may not be Independent Spring 2004 ECE39EE H H quot9 CWWH M Lecture i Side is Spring 2004 ECE39EE H H quot9 WWW Lecture i Slide i4 Source Entropy Source Entropy The average selfinformation of such a lengthn if X are iid independent amp identically diStribUtedL sequence is witthenoting a generic random variable as X G 7ZZZPIX104X2042X 04 Gi 6sz atsuXi 64 x1 121 11 lo PrXa X aXna logPrXiotX2otXor g quot 2 quot l 2 i 7 m P X Per P X quot1 PrX The entropy of the source per symbol is defined as r aquot aquot r a Egg aquot HS hm Git iniP X alogPrX 04 new n 11 m HS1im Gquot eZPmX ilogPrX i In the lack of complete knowledge of the experiment H n i assumpiions are P en made to faCiiitate entropy lfthe condition of iid is assumed ratherthan a given caiCUiationi eQi quotdi Markovi fact then the above H6 is called 1st order entropy Spring 2004 ECEEEM H H quot9 CWWH W Lecture i Side is Spring 2004 ECEEEM H H quot9 W EH W Lecture i Slide we ECE 8833 Lecture 10 Feb 17 2003 Queuing amp QOS Classic Queuing First In First Out FirstIn FirstOut Packets arrive and are placed at tail of queue Packets are removed from head of queue Packets are dropped if insuf cient space FIFO problems packet size TWO streams of packets into same queue Stream 1 100 pps 100 bytes each 0 10000 bps throughput Stream 2 10pps 1000 bytes each 0 10000 bps throughput If link is nearly saturated only Stream 1 will pass stream 2 takes 10 times as much space in queue This is usually handled by preallocating sized buffers and queuing pointers to the buffers ie 20 buffers for packets gt1024 bytes 4O buffers for packets 512ltsizelt1024 80 buffers for packets 256ltsizelt512 FIFO Problems Greedy source In a saturated network 0 Stream 1 100 bytespacket lpacketsecond Stream 2 100 bytespacket 100 packetssecond Stream 2 will get higher throughput Stream 2 will get a higher percentage of packets through Note that this stream scenario is typical of the TCP vs UDP sharing situation and aggravates this problem FIFO Advantage It is gtkvery lt simple to implement Random Early Detection 0 Simple variation of FIFO Situation Many Many TCP ows through a FIFO Queue Queue nearly full running at capacity Sudden traf c burst causes losses everyone detects congestion and slow down to 12 speed Link is now running at 12 capacity TCP rampsup til queue is full again Sawtooth behavior between 50 100 utilization Called Global synchronization RED 2 P RED Randomly drop packets when queue is near full Based near full on running average of queue usage Helps global synchronization Also helps with packet size and rate fairness Finding good values for threshold and drop probability can be dif cult Bad values can be worst than FIFO WRED P Weighted Random Early Detection WRED Use different drop probability curves for different types of traf c IE RED curve for TCP GREEN curve for UDP or by port numberapplication etc Used to give some applications priority Multiple Queues Three Sections 1 SortClassify 2 Multiple Queues 3 Selection Process Queue Server Sorting Based on Physical input link into system router Packet Header information Router State information packet header Note that frequently the number of queues available forces multiple sorting criteria to map to one queue ie all RSVP traf c to one queue Queue Server Many Possible Different Algorithms Strict Priority Weighted Round Robin Weighted Fair Queuing Strict Priority Order Queues by priority Take packet from high priority nonempty queue Starvation is likely unless care is taken Round Robin Take one packet from each queue in turn Weighted version Assign a number of packets to take from each queue in turn Bounded Delay Session Initiation Protocol ECE 8833 March 10 SIP Functions User Location User Availability User capability Session Setup Session Managrnent SIP Not an application or a transport VoIP UDPRTP Meant to be something like Hypertext transfer protocol HTTP SIP Partners Session Descriptor Protocol SDP RealTime Protocol RTP MEGACO SIP Proxy Servers Anchor points for SIP activity mobility lockup Similarities to an Email server Identi ed by a name SIP URI Sipuserh0stnarneccrn Sipsuserh0stnarneccrn secure
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'