Class Note for EECS 563 with Professor Frost at KU (3)
Class Note for EECS 563 with Professor Frost at KU (3)
Popular in Course
Popular in Department
This 28 page Class Notes was uploaded by an elite notetaker on Friday February 6, 2015. The Class Notes belongs to a course at Kansas taught by a professor in Fall. Since its upload, it has received 38 views.
Reviews for Class Note for EECS 563 with Professor Frost at KU (3)
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: 02/06/15
Network Traffic 6 Victor S Frost Dan F Servey Distinguished Professor Electrical Engineering and Computer Science University of Kansas 2335 Irving Hill Dr Lawrence Kansas 66045 Phone 785 8644833 FAX785 8647789 email frosteecskuedu httpwwwittckuedu Section 471 572 Traffic 1 Traffic Characterization I Goals to gt Understand the nature of what is transported over communications networks gt Use understanding to improve network design I Traffic Characterization describes the user demands for network resources gt HOW often a customer Requests a web page Down loads an MP3 Makes a phone call gt Size length Web page Song Phone call Traffic 2 Voice Traffic Aggregate Traffic I Requests for network resources from a large population of users 1 Large Population Traffic 3 Voice Traffic Aggregate Traffic I Arrival Rate gt Number of requests time unit Calls sec I Arrival rate A I Holdmg Trrne length of tune the request will use the network resources gt Min call gt secpacket Average Holdm g sze Th Traffic 4 Voice Traffic Aggregate Traffic I Traffic Intensity load gt Product of the average holding tilne and the arrival rate Traf c Intensity p itle I Units of Traffic Intensity gt Erlangs gt CCS 1 CCS is 100 call sechr gt 1 Erlang 36 CCS I Traffic intensity is specified for the 39Busy Hour39 Traffic 5 Voice Traffic Aggregate Traffic I A telephone line busy 100 of the time 1 Erlang I A telephone busy 6 min hour is how much traffic gt 01 Erlang gt 100 telephones busy 10 of the time is how much traffic gt 10 Erlangs Traffic 6 Voice Traffic Aggregate Traffic I A telephone busy 100 seC Centi call Seconds CCS per hour 1 CCS I Arrival rate 1 call hour I Average holding time 100 seC I 80 gt 100 sec call 1 hour 36005ec1call hour 136 Erlangs 1 CCS I 36 CCS 1 Erlang Traffic 7 Voice Traffic Aggregate Traffic I Traffic is Random gt Holding time gt lnterarrival time time between calls I Common assumptions for probability density function pdf for gt Holding time N exponential gt lnterarrival time N exponential Sec 0n471andA11 Traffic Voice Traffic Aggregate Traffic Interarrival Histogram Percent Ilems ken 539 Distribulfon 1315 39 39 39 12 05417 1035833 98625 8 766667 7 670833 6575 5479167 H 4333333 1quot a 2875 H 2191657 1 095533 00003591753 1318353 2535945 n efrig gglq me 5271133 5588725 7 90632 Traffic 9 Vonce Traffic Aggregate Traffic PTh ltt 1 239Mf0rt gt Ound 0f0rt lt0 1 Th u PT1 ltt 13 tf07 t gt0 and 0f0rt lt 0 T I Interurrivultime Traffic Voice Traf c Aggregate Traf c T Time hb bl Time rmmc 11 Voice Traf c Aggregate TrafficDiurnal Variation Call Patterns UNI11Dynamxc Nonrhxerachxal routing Volume of calls 3 am 12 noon 5 pm Time of Day mm 12 Voice Traffic Individual voice source I Speech inactivity factor Talkspun Takspurt Talkspurt JIIll Silence Silence me mm 13 Voice Traffic Individual voice source I Talkspurt duration gt Random gt Average duration gt 0350 s to 13 s gt Exponentially distributed I Silence period gt Random gt Average duration gt 585 to 165 gt Exponentially distributed I Speech activity factor 035 to 436 Section 5 5 2 mm 14 Voice Traffic Individual voice source I Digital Speech Interpolation D81 6mm gt Uses quotsilence detection gt Multiplex at the talkspurt level gt View as call set up at talkspurt level gt NDoubles the capacity gt Analog version called quotTime Assignment and Speech Interpolation TASI gt Packet Voice with silence detection effectively does DSI Traffic 15 Voice Traffic Individual voice source I Signal redundancies gtVoice coding I Pulse code modulation 3711 PCM 8bits sample 8000 samples sec 64kbs I Adaptive Differential PCM 32kb s I Linear Predictive 24 to 16 kb 5 I For Voice over IP rate lt 8kb s gt G7231 is emerging as a popular coding choice G723 is an algorithm for compressed digital audio over telephone lines Traffic 16 Comparison of popular CODECs Compression Compressed rate Required CPU Resultant voice Added delay scheme Kbps resources quality 6711 PCM 64 no compression Not required Excellent NA 6723 M PMLQ 6453 Moderate Good 64 High Fair 53 6726 ADPCM 403224 Low 600d 40 Very low Fair 24 6728 LDCELP 16 Very high 600d Low 6729 CSACELP 8 High 600d Low There is no quotright CODECquot The choice of what compression scheme to use depends on what parameters are more important for a specific installation In practice G723 and G729 are more popular that G726 and G728 For details of other VolP Codecs see Tram httpwwwzylraxcomtechprotocolsvoipirateshlIn Voice Traffic Individual voice source I Example How many calls can be supported on a system with the following parameters gt TDM gt Coding rate voice Channel ADPCM 32 Kbs gt DSI gt Line rate 1536 Mbs note a T1 D81 line is 1544 Mbs I Number of ADPCM Channels 1563 Mbs32 Kbs 48 I With DSI you get 2 calls Channel 96 Traffic 18 Voice Traffic Packet Voice I Example Parameters for a packet voice system gt 1 source gt Sample rate 8000 samples sec ITU 3711 gt 8 bits sample 1 byte sample gt 8 mspacket Critical parameter gt Link rate 10 Mbs gt bytes packet 8ms packet8000 bytes sec 64 Bytes assuming no overhead bytes gt Holding time packet 64 bytes packet8bits byte 10 Mb s 512us Traffic 19 Voice Traffic Packet Voice I 512 us Transmit at a Constant Bit Rate CBR l Il Il I l smS l 8ms gt I time I 512 us Receive with variable interpacket arrival times l Il Il I X ms 39 tilne X not equal 8ms because of network delays If X is too big packet may arrive too late for play out Traffic 20 Voice Traffic Packet Voice Perfect Multiplexing of N VolP sources I 512 us I 1 2 I O O O I N 1 I I 8 ms tune Traffic 21 Voice Traffic Packet Voice I Packet voice looks like a steady ow or Constant Bit Rate CBR traffic I However voice can be Variable Bit Rate or VBR gt quotsilence detection gt Variable rate coding I Problem After going through the network the packets will not arrive equally spaced in time Thus playback of packet voice must deal with variable network delays Traffic 22 Voice Traffic Packet Voice I Assume network dela is uniformly distributed between 5 ms 75 ms gt Same as having a fixed propagation dela of 25 ms with a random network delay un39 ormly distributed between 0 ms 50 ms I Note receiver will run out of bytes to play back after 8 ms I Solution gt Buffer 50 ms or 8 packets or 28 Kbits gt Worst case receiver will run out of data just as a new packet arrives Traffic 23 Voice Traffic Packet Voice I New problem networks delays are unknown and maybe unbounded I A voice packet may arrive at 85 ms and be too late to be played back gt Late packets are dropped gt Last packet may be played out in dead tilne I Packet voice video schemes must be able to deal with variable delay and packet loss Traffic 24 30 ms 2 byles Voice Traffic Packet Voice lTUrT Recommendation G11470nway transmission time May 2003 G 7251 is a a voic e co ding n 4nz Frame 5 792 5 4 5 standard linear predic on c ompression U algorithm IP header um RIP Payload pquot quot 39 fmam 171 kbs 20 a 12 40 byxes 24 bytes Cumpressed Comprasea header F Gamer1m 51mm 75 kbs m cm 39 quotmad RFC 2508 4 24 mm Mama Manama Ammanquot mmmrm Quhvansemce pmmg Tvamc 25 Katsnyushx hda Ken Kwahm mm Takmz and Yup 0 IEEE Cammumcanans MagnumApn12 mu i 4 w 9H m in an r 3 E m 7m 5m 5n n mm mm mm ADD Sun I uojm Maumamamym Tvamc 26 VoIP Factors in End to End Delay lAssumption maximum delay from earto ear needs to be on the order of 200 300 ms 53 gt ITU 6114 quot e lt 150 ms acceptable for most applications 7 150ms 400 ms acceptable for international e gt 400 ms unacceptable Nuwmx39 mm l l E m gt MW W mn 39Rec wer quot mete ML 27 Frorn ht www rotocolscom a ers V01 2mm Tyan VOIP Factors in End to End Delay I Example Delay Budget depends on assumptions gt Formation of VoIP packet at TX N 30 ms 20ms of voice packet is default for Cisco 7960 router gt Other VoIP packet processing 70 ms see httpwwwmavaraucbrpdfvoippdf gt Propagation 20 ms gt Extraction of VoIP packet at Rec 30 ms gt Jitter Buffer N 100 ms Compensates for Variable network delay gt Total 250 ms m hup www lightmadmg Comdocumentasp7s e1ightreadmgi zdociid538644 zpageinumber6 Tvamc 28 Data Traffic General Characteristics I Highly variable I Not well known lLikely to change as new services and applications evolve Data Traffic General Characteristics I Highly bursty where one definition of burstyness is Peak rate Burstyness Average rate Traffic 30 Data Traffic General Characteristics Example During a typical remote login connection over a 192kb s modem a user types at a rate of 1 symbol sec or 8 bits sec and then transfers a 100 kbyte file Assume the total holding time of the connection is 10 min What is the burstyness of this data session Traffic 31 Data Traffic General Characteristics The tilne to transfer the file is 800000 bits 19200 b s 41 sec 80 for 600 4lsec 559 sec the data rate is 8 bits sec or 4472 bits were transferred in 559 sec Thus in 600 sec 4472 800000 bits were transferred yielding a average rate of 804472 bits 600 sec 1340 bitssec The peak rate was 192 Kb s so the burstyness for this data session was 192001340 143 Traffic 32 Data Traffic General Characteristics Call AIDW1 T r 1 T I Sess1or1 Interarrivals N i Call 1 I I Dma on H H Sess1or1 Duration 1 r Voll i quotm packet Packet Interarrivals Auivals w Voll Packet i ng39l j l Packet Lengths Tmit 38 Data Traffic General Characteristics Asymmetric Nature of Interactive Traffic Think Time User Burst User Burst Idle Timk Idle Time Computer Burst Computer Burst This Asymmetric property has lead to asymmetric services Tra fi 34 Data Traffic General Characteristics I In Time Division Multiplexing TDM user must wait for turn to use link I Statistical Multiplexing Stat Mux gt Note high burstness leads to quotlongquot idle times gt By transmitting the bursts39 on demand the link can be efficiently shared gt To help insure fairness break the burst39 into packets and transmit on a packet basis Traffic 35 TDM vs Stat Mux Userl m Dwell time one time slot o o o IIID Server N USE Server Traffic 36 Data Traffic General Characteristics I Element length gt Message gt Packet gt Cell I Arrival rate gt Message sec gt Packets sec gt Cells sec Traffic 37 Data Traffic General Characteristics I Traffic intensity lt 1 with one server p 1 Th where Average Packet Length in Bits L h Link Capacity in Bits sec C Average Packet Length in Bits L Link Capacity in Bits Sec C Traffic 38 Data Traffic General Characteristics I Smndard Assumptions gt Message length has an exponential pdf gt Interan39ival time has an exponential pdf E t Data was taken from speclal tmces m httpl lwwwnlanrnet Data was captured at thelntemet Upllnk of the Unlverslty of Auckland by the Wand Reseamh group m the year 2000 The tap was lnstalled on an oce llnk Tram 39 Figure 2 Complementary Cumulative Distribution Function 0 packet interarrival times over two backbone networks For 0048 link traces on a January 2003 backbone 1 and b January 2004 backbone 2 as well as for c the BCpAu989 1989 Belcore trace the yaxis is plotted in logaritth scale We can approximate the distributions 0 0048 traces with an n mm mm m m I M H m exponential distribution L straight line in loglinear scale but t 1 the BCpAu989 data set clearly gm 1 deviates from the exponential E t 3 distribution goquot t 1 i t I K 1 i l 1 1 t t From LongRange Dependence Ten Vears orntemet m m m a m M Traf c Modeling EEE lntemet Compound Sept out mm m Sprmt H551 5m D53 Fasl hemet m I KU ITTC has collected WW mu Mb mm mm aggregate traffic data from News Sun ower Datavision ex smmam came Madam Note Came System15 a DOCS1S System 115mg C sco MC16 Cards T1115 a110W5f0r1 downstream Channe1and 6 upstream Channe15 Traf c 4 1 From the Internet into Datavision From Datavision out to the Intemet Mean 8876 Mbs Mean 5133 Mbs Maximum 18952 Mbs Maximum 12093 Mbs Traf c Flow 9 oiosao itInDCtEt on For H15 H Mbs Man Feb 10 000000 2000 da gs v0 00 11 000000 2000 Man Feb 10 000000 2000 da gs v0 00 11 000000 2000 From the Internet into DataVision From DataVision out to the Internet Mean 8555 Mbs Mean 4343 Mbs Maximum 21597 Mbs Maximum 12093 Mbs Traf c 43 Data Traffic Conclusions I Very bursty I Problems with traffic modeling gt Rapidly evolving applications gt Complex network interactions I Issues gt Do models match quotrealquot traffic flows gt Are the performance models based on specific traffic assumption robust Traf c 44 Video Analog video I Bandwidth 4 Mhz I Uncompressed rate 64 Mb s I Components of the signal gt Luminance gt Chrominance gt Audio gt Synchronization Traffic 45 Digital Video JPEG I quotJoint Photographic Expert Groupquot Voted as international standard in 1992 I Suitable for color and grayscale images eg satellite medical applications I Targeted for still images I Capable of reducing continuous true color or gray scale images to less than 5 of their original size I JPEG GIF and MPEG define the compression as well as the Video data format Section 12 3 Traffic Digital Video MPEG I Moving Pictures Experts Group I Compresses moving pictures taking advantage of frametoframe redundancies I MPEG Initial Target VHS quality on a CDROM 320 x 240 CD audio 15 Mbits sec Traffic 47 Digital Video MPEG I Converts a sequence of frames into a compressed format of three frame types I I Frames intrapicture I P frames predicted picture I B frames bidirectional predicted picture Traffic 48 Digital Video MPEG Exploits frame to frame redundancies Traffic 49 A 1000000 52 Frame frame a 800 000 Sizesfor P frame CD 600000 B frame actzon GE 400000 video 3 200000 w L 0 0 200 400 600 800 100C Frame number 1000000 32 800 000 Frame sizes i I frame Each frame would 8 i P frame be trans orted a 600000 for talking B frame Flt l 395 head video 5mg m 1P e 3 400000 7 vaMMK mvw Adwmk packets E 200000 28m LL hwmwvu F r i e0 reams 0 n gtM w 4ywmip overATM Steven enngen O 200 400 600 800 1000 etai iEEE Muitimedia Frame number 1998 50 Traffic MP3 MPEG Layer 3 Audio MPEG specifies a family of three audio coding schemes Layer1 2 3 Each Layer has and increasing encoder complexity and performance sound quality per bitrate l The three codecs are compatible in a hierarchical way ie a LayerN decoder is able to decode bit stream data encoded in LayerN and all Layers below N l The MP3 compression algorithm is based on a complicated psycho acoustic model I The majority of the files available on the Internet are encoded in 128 kbitss stereo l A high quality file is 12 times smaller than the original I CDs can be created that contain over 160 songs and can play for over 14 hours on a PC I Music can be efficiently stored on a hard disk and then directly played from there Traffic 51 Digital Video MPEG I Compression ranges gt 30ml gt 50to 1 I MPEG is evolving gt MPEG 1 gt MPEG 2 gt MPEG 4 gt MPEG 7 Traffic 52 Digital Video MPEG4 I Audio Video coding for quotlow bit rate channels gt Internet gt Mobile applications I MPEG 4 is a significant change from MPEG 2 I Scalability is a key feature of MPEG 4 I MPEG 4 contains a Intellectual Property rights IPR management infrastructure Traffic 53 Digital Video MPEG4 I Object based AudioVisual objects AVO I AVO are described mathematically and given a position in 2D or 3D space I Viewer can change vantage point and update calculations done locally I No distinction between mzturul and synthetic AVOs treats two in an integrated fashion I Each AVO is represented separately and becomes the basis for an independent stream I Each AVO is reusable with the capability to incorporate onthe y elements under application control I Content transport with QoS for each component Traffic 54 Conclusions I Network traffic defines the demands for network resources I Network traffic is dynamic gt Changes with the deployment of new application gt Time of day I Models for network traffic are continuing to evolve Traffic 55
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'