`
`1
`Special Issue
`
`CNETDP 26 (824191. 2,3) S53-S146 (19§4)
`
`!SSN 0189-17552
`
`
`
`A Supplement to
`
`Volume 25 (1994)
`
`2—-DEC--1984 BLDscBE§;§“?§§*
`CEMPUTER NETHURKS FQHD ISBN SYSTEHS
`_
`11111111
`man
`
`
`339.117UUO 93% 33? /2/3
`
`#99‘
`
`
`
`I
`
`1
`
`1
`E
`1
`2
`
`
`
`A Selection of Papers
`Presented at the Network
`Services Confe-rence1994
`
`
`
`ELSEVIER
`
`RPX Exhibit 1121
`RPX V. DAE
`RPX v. DAE
`A Sepgeéemeni to Commie? Nemserks am“; 18%: Syséems Cemmsezf by RARE em EARN
`
`;
`g
`
`RPX Exhibit 1121
`
`
`
`
`
`V COMPUTER
`NETWORKS
`for RESEARCH in EUROPE
`
`A Suppiemerzt to
`
`COMPUTER
`ggdsrwonxs
`ISDN SYSTEMS
`
`Vaiume 26
`1 994
`
`
`
`Amsterdam — Lausanne —— New ‘/ark — Oxford —- Shannan — Tokyo
`
`
`
`
`
`Abstraciedfindexed in:
`A85/Inform
`ACM Computing Reviews
`Cambridge Scientrféc Abszrams
`Computer Abszracts
`
`Computer Lrterazure index
`Engineering tndexicornpendex
`WSFEC lntormatkon Services
`Lébrary and mformazion Science Abstracts (USA)
`Sociological Abstracts
`Zenrrajblatt fur Mathematék
`
`@ The paper used in this msbéacarson meets me requirements at ANSUNISO 239.48»3992 (Permanence of Paper).
`© 1994, Ersevier Science Publishers B.V. All rights reserved
`No part of this publication may be reproduced, stored in a remevai syszem or transmfszed in any form or by any means, eiectronéc, mechanécaf,
`photocopying, recarding GI osherwise. wéthaut me prior permission of me pubiisher, Elsevéer Science B.V., Copyngh: & Permisséorzs Department,
`PO. Box 52%, 1008 AM Amsterdam. Tne Nemeriands.
`
`irrevocable and exciusive authprézatéon of {he pubtésher to collect any sums or conszaeraréons for copying G!‘ reprcduciion payatée by third parties
`{as mentioned in artiste 17 par:-graph 2 of the Dutch Copyright Act of 1912 am: in the Royal Decree of June 213, 1974 (S. 351) pursuant to article
`15b of the Dutch Copyright Act of 3932} arwsr £9 ass: 6:‘: or our of Court in connection therewith.
`Special regu/at/Lens for readers in the 2.18% - Tms gournal has been regésxered wém the Copyright Crearance Cerzzer, inc. Consent is given for copy-
`ing at‘ artistes far persona! or mzemaf use, or for the persenaé use of specific ciéents. This consent is given on the condition that me capier pays
`through the Center the per-copy fee stated it’! me code on {he firs! page of each arsrde Iar copying beyond that permitted by Sections 307 or ‘H38 of
`me U.S. Cepyrighr Law. The apprcpréare fee sracsurcf be forwarded mm a cam: cf me first page of the artecie to the Copyright Chaaranoe Center.
`mc., 222 Rasewocrs Drive, Barwers, MA CHQ23, USA. £7 rte cede appears in an amcée. the author has not given broad consent to copy and permis-
`sron is copy must be obiamed %érectt-,2 ifrem the author. We fee indecazed on the firs: page cf an arzicia wit! appiy retroactivery in all articles pub-
`iéshed m me journal. regarritess 0! the year of pumécaiicn. This consent does not extend :0 other kinds :2! copying, such as {at genera! distribution.
`resaie, advertising and prometron purposes. er for crearing new corrective works. Specéat wrmen permission mus: be obtamed from {he publisher
`for such oopyéng.
`
`No responsubeusy rs assumeé by the ca.-béésmar ?or any mgury and/at damage to persons or property as a matter of pmduczs riabiféty, negiigertce or
`otherwise. or mam any use or operation of any methods, products, értstruczioras or édeas contamecs an the maIen’a§ herein.
`Arthough ail arjverzzssrrg maiersaé as expected £0 conform to efihrcai standards. mctusécn In this pubiicazéon does not conszitute a guaramee or
`endorsement of me quaiity or vaiu-3 0! such faroduct or 0:! the ciarms made or R by ets marnzfacturer
`Primed an the Nezhereands
`
`G359-?552/94/$07.06
`
`
`
`Computer Networks and lSDN Systems -— Vol. 26 Suppi. 2,3
`
`
`Abstractedflndexed in: ABE inform, ACM Computing Reviews, Cambridge Scientific Abstracts,
`Computer Abstracts, Computer Literature
`index, Engineering lrzdexfliompendex, ENSPEC
`information Services, Library and lnformaticn Science Abstracts ELESA). Socioiogical Abstracts.
`Zentralblatt ffir Mathernatiic.
`
`Special issue: A Selection of Papers from the Network Services Conference 1994
`
`Contents
`
`Editorial
`
`Tom Baker, inke Brfining, Lemar Kiein and Michaei Lenz
`Starting a Web site: issues of quality control
`
`Hermarzn Maurer and Klaus Scrzmaranz
`
`JUGS ~ The next generation in electronic journal pubiishing
`
`Jon Crowcroft and Gordon Joiy
`Touring and navigating a giobal learning environment — the Web
`
`Miriam Farber
`
`The quality of information on internet
`
`Christopher Douiney
`lBSS ONLINE
`
`Mary Keefer
`A new approach to digitai resource development {a new method for marketing
`digital technology)
`
`Anne M. Mumford
`
`Supporting multimedia in UK higher education
`
`853
`
`855
`
`883
`
`871
`
`875
`
`879
`
`885
`
`893
`
`Pierre Ageron, Francois Besson, Patrick Desfarges
`images database management system: a “semer~ciient-producer" system on a
`local network and on the Internet
`
`3101
`
`Manfred A. Bogen, Harisiudwig Hausen, Flainer Worst
`Handling of Q08 characteristics
`
`zfike Gréinfunc-’
`
`' Public computer systems — A chaiienge for orgartizationat learning
`
`Serge Desararmo and Ferdi Put
`Coordinating group communication
`
`Carla Gaibisso, Giorgio Gambosi, Maurizio Lancia, Maurizio Vitals
`Muitimedia conferencing on packet switched netvvorks: Testing and evaiuation
`
`S10?‘
`
`S2119
`
`3129
`
`8139
`
`lllllllllll llll
`
`
`
`l
`
`ll 1
`
`
`
`III lllllll
`
`
`
`
`
`lllllllllllll ll
`
`
`
`
`B169-75S2{199=i»}2i$:s2/‘S3; ‘i -C
`
`
`
`COMPUTER NETWORKS
`and ISDN SYSTEMS
`
`The Inzernationai eournai at computer and Teiecommunications
`Networking
`
`Scope
`COMPUTER NETWORKS and %SDN SYSTEMS is an
`iniernaiioriai iourrrai providing a pubiicaiion vehicie for
`compiete coverage of eii topics of iriieresi to {hose
`irweiired in the area. The audience irrciedes mari-
`egers and operators of nerwsrirs as weii as designers
`and impiemenizors in addition to reguiaiory iawyers,
`ecorromisis, policy anaiysis, and so-cioiogisis speciai—
`izéng in rise area. The Ediioriai Beard wiii consider for
`ouoiication any mereriai
`trier‘
`is of irrieresi to mess
`groups.
`
`Subject coverage
`Maieriei on air’ aspecrs ei ihe design, impiemeniaiiori,
`use and management or «semester and zeiecommurw
`caiion rieiworks, oommuriicaiéorr sub-sysrems. and
`inregrared
`services
`digiiei
`networks.
`Sgoecéiicaiiy
`irrciuded are new-rechnicai topics such as economics,
`iegai and regulatory issues, errci social impact.
`Types of contributions considered
`The primary purpsse of ihe ieurnai is is oubiish cem-
`pieie papers covering a specific roeic or project in suf-
`ficient detaii and depih is be oi praciicai use is misr-
`ested readers. In addiiieri is iu§i—ierig£ri papers, sheri
`noies reporting on a smaii oroieci or giving an interim
`report on a oroieci wiii arse be considered. iris jour-
`nai is meant to be a compieie source or’ information
`on an aspects of compuier and ieiecommunicaiion
`networks and wiii aiso irieiucie news items and adver-
`tisements in addition to the reguiar eeriiribuiions dis-
`cusses above.
`
`Submission of materiai
`3% is preferred that oorrroieie rrrarrusrsripzs be submiiied
`to one oi iiwe Editors. menéioneo
`the mesrheacf {err
`the iérsi inside cover page ei this issue).
`Review procedure
`Aii papers and error: notes are suomirted to expert rei-
`erees ior review. _Ԥ'i'ieir reports are iuiiy considered by
`the Ediroriai Board is seieciing maieréai for pumice-
`iiori. The names oi specific reierees are never
`ciivuiged to the author; hewever, same or irie review
`comrrieriizs may be prcnrided to assisi the awzrror in
`revising his paper. Leiters are reviewer: by an Editor
`for appropriateness and cement. ii ii is necessary to
`shorter: or otrierwise edit maieriai submitted, iriis fact
`wiii cieariy be indicared in {he eubiisired form.
`Author's benefits
`
`1. 50 sfiprinis of each csniribution free of charge.
`2. 30% discount on aii Eisevier Science Pubiishere
`books {order forms wiii be sent iogeiher with the
`prosis}.
`
`-
`Subscription information
`Computer Networks and iSDN Sysiems {iSSN 3169-7552}.
`For 2994 voiume 26 23 scrieduied for pubiicarion A com»
`
`and SEN
`subscription to Csmzauier Networks
`Sines‘
`Sysiems ans‘ Computer Networks. for Research in Europe
`(suppiemeni to voiume 26} £4 issues‘; a: reduced rate is
`avaiiebie. Subscription prices are avaiiabie upsri request
`from the i3Ubi§S3‘rEI_ Subscriptions are accepted on a prepaid
`basis oniy and are entered on a caiendar year basis. issues
`are sent by surface maii exept to the foiiowing countries
`where air deiivery via SAL is ensured: Argeniina, Australia,
`Brazii. Canada, Hong Kong, India. israei, Japan, Maiaysia.
`Mexico, New Zeeéand, Paiesien, PR China, Singapsre,
`South Africa, Souih Korea. iaiwen, Thaiiand. USA. Fer aii
`other countries eirrneii rates are avaiiabie upon requesi.
`Ciaims far missing issues must be made within Six months or
`our oubiication imaiiirig} daze. Piease address air your
`requests regarding orders
`subscription cguerres
`to:
`Eisevier Science, Journai Department, PO. Box 2M, 7008
`AE Amsterdam, The Neiheriarios, Te%.: 3i—20—48S3542, fax:
`31 -204853598.
`
`US mailing notice -~» Computer Networks and ISDN
`Systems £0769-7552} is oubéished merririigr by Eisevier
`Science {Mcienweri T, Poszeus 2i 1,
`€500 AE Amsterdam}.
`Annuai subscription price as rhe USA U535"/’9.0Q {USS prrce
`vaiid in North, Cenzrai and South America Gniyj inciudmg air
`speed deiivery. Appiicarion to maxi at second crass postage
`rate is pending at Jamarca, N‘! ‘r $431.
`USA PGSTMASTERS: Send address changes io Computer
`Networks and iSDN Sysierrrs, Pubiicatiori Expediting,
`inc..
`2200 Meacharn Avenue. Eimorrr, NY H003. Air ireighi and
`maiiing in the USA by Pubiicaziori Expediiing.
`
`© 1994, Eisevier Science B.V. Mi rights reserved
`
`No part or this pebiication may be reproduced. sroreo‘ ii’? a rernevai
`syséem or rransrmiied in any form or by any means. eieczronsc,
`IT'}eCi'!3T!éCai. phoiocosyirrg. recordrrrg er otherwise, wizmm the gamer
`permission oi
`ihe pubissher. Eiser.-re: Science BM, Copyright
`is
`Permissions Deparimerrz, P O. Box 52?, W90 AM Amsierofam.
`Nezheréarids.
`
`Specie! reguianons fer aumors - Upon acceptance or en amcie by
`the iournai, the aU!?70¥’{'S} wiii be asked to trarisier copyright or the
`ariicie to me puoiisher. This trarasrer wrii ensure the ii.-ides: possibie
`dissemrnarien or iswfenrzairen
`
`Specrai reguiarrons for readers in me use — Thrs geurnai has been
`regzstered with the Cooyrigrit Ciearance Center, éric. Consent IS given
`for seoyirrg of armies ior persona: or mzernai use, or for me persona?
`use of specaiic céierrrs. ms coriserii is given on ire condiizeo mat the
`cogziier page through the carrier the Qer-cqoy fee stated in the code or:
`me first page of each arzecie fer cooyzng beyond mat perrnsztedé by
`Sections 107 or 108 or the iJ.S Copyrsghz Law. The appropriate iee
`shouici be forwarded with a copy oi the firs: page si me arricie 26 the
`copyright Clearance Cemer. ii"iC.. 222 Rosewood Drive. Danvers, MA
`01923, USA.
`if no code appears !€'1 an arzscée. zhe author has not
`given broad consent
`is copy and D€!|'¥'iiSS!Q!‘i
`£0 cepy mus: be
`oozaened direcriy from me author, The fee irieicaied on the first page
`oi as am-are m ems issue wiii a_og2;i'y rerroacixueiy ro air amcies pub
`risheci rn the yourriai. rergardiess of me year at pueiicaéien Thrs cenw
`sen: does not ezzend to omer kinds or copying Such as fer genera:
`drsrriouisorr, resaie. advertising and eromoiion purposes. or tor creat-
`mg new corrective works Specie: wrrzreri permsssson mus: be as
`ierneo from me puoiisher isr such copying
`
`No responsibziiry i3 assumed by the pubéssher for any mgury and/Or
`damage to persons or property as a matter of products izabriity. siege-
`genes or otherwise, or from any use or oeerazren oi any meiirecis.
`products. instructions or roses contained in the materrai herein.
`ixrihougia aii edverzzsing rrraieriai
`is exwzed to conform to ethics:
`siancfards, inciusion ii’? ms puoiicazion does not constriuie a guaran-
`-tee or endorsernenz oi the Quaiéiy or vaiue of such precise: or or me
`ciazms made or ii by srs manofacrurer
`
`T"-9 paper user: 4: ms oueizcarw meets the reswamer-rs o‘ .-ibis! N‘-‘SO
`23$.-48— ? 992 ~‘Derrnanenr;e oi Paper: .
`
`
`
`Computer Networks and ISDN Systems 26 (Suppl. 2 double issue) (1994) S139-S146
`
`Multimedia Conferencing on Packet Switched Networks:
`
`Testing and Evaluation
`
`Carlo Gaibisso, Giorgio Gambosi, Maurizio Lancia, Maurizio Vitale*
`Irtituta di Analiri dei Sistemi ea’ Irtfvrmatica, Crmsiglio Nazionale delle Ricerche, viale Manzoni 30, I-00185 Roma, Italy.
`
`Abstract
`
`This paper briefly reports on the results of some testing activities performed at IASI on the efficiency of multimedia
`conferencing on both local and wide-area packet switched networks.
`in terms of rate of
`The aim of the testing was to study under which hypothesis it was possible to obtain acceptable -
`information exchanged - multimedia conferencing sessions. A specific goal was to verify the possibility of obtaining an
`acceptable quality by using widely available, cheap resources.
`The tests performed have been carried out on an Ethernet LAN, connected to Internet by a 64Kbps link, using TCP/
`IP with multicast extension as transport protocol, in order to avoid IP datagrams multiplication when a multimedia con-
`ference involves more than two participants.
`
`Keywords.‘ Video conference, Audio conference, H.261, MPEG, Traffic measurement
`
`1. Introduction
`
`This paper briefly reports on the results of
`some testing activities concerning multimedia
`conferencing on packet switched networks. Our
`aim is to show that multimedia conferencingof
`acceptable quality is possible using widely avail-
`able cheap resources, such as desktop computers
`and packet switched networks without guaranteed
`bandwidth (Ethernet, Internet),
`instead of the
`
`powerful workstations and circuit switched com-
`munication which usually require guaranteed
`bandwidth. Our tests have been carried out on an
`
`Ethernet technology based LAN, connected to the
`Internet by a 64Kbps link. The transport
`level
`protocol has been TCP/IP with multicast exten-
`sion, which avoids IP datagrams multiplication
`whenever a multimedia conference involves more
`
`than two participants.
`
`*Corresponding authors:
`Email: gaibisso@iasi.rm.cnr.it, gambosi@mat.utovrm.it,
`lancia@iasi.rm.cnr.it, vitale@iasi.rm.cnr.it
`
`Routers supporting multicast transmission have
`been implemented by establishing virtual point—to—
`point connections (tunnels) between remote sys-
`tems (the same technical solution adopted in
`setting up Mbone [2] virtual connections). Audio
`and video conferencing have been tested separate-
`ly to evaluate how the efficiency of transmission
`is influenced by the nature of the transmitted
`data.
`In this respect different audio and video
`conferencing tools have been used with either
`standard or extensively used compression schemes
`implemented. Most of them are in the public do-
`main and are available on different operating plat-
`forms.
`
`With only one exception, such compression
`schemes are implemented via software,
`thus re-
`quiring only simple boards for data acquisition.
`As a consequence, our tests also take into account
`the time spent
`in encoding/decoding.
`
`0169-7552/94/$07.00 © 1994 - Elsevier Science B.V. All rights reserved
`
`
`
`
`
`S140
`
`Carlo Gaibisso, Giorgio Gambosi, Maurizio Lancia, Maurizio Vitale /
`Multimedia Conferencing on Packet Switched Networks: Testing and Evaluation
`
`All tools have been tested on several architec-
`
`tures, with different performance capabilities,
`ranging from Sun Sparc Station 1+ and Sparc
`Station 2,
`to Iris Indigo class Silicon Graphics
`workstations, to more powerful machines such as
`Sun Sparc Station 10 and DEC 3000 AXP.
`
`2. Traffic measurement
`
`issues related to audio and
`As already stated,
`video conference are considered separately.
`
`2.] Audio conferencing
`
`The performances of two different tools for
`audio conferencing have been investigated: Visu-
`al Audio Tool [7] (VAT) and Network Voice Ter-
`minal
`[3]
`(NEVOT). Our measurements include
`the CPU time devoted to encoding/decoding, the
`number of packets generated‘ and the playout de-
`lay. Moreover,
`the amount of network traffic,
`in
`terms of the number of bits passing through the
`interface between the access router and the link
`
`to the Internet, has been evaluated, since it is in-
`
`dicative of the bandwidth requirements. Finally,
`the level of intelligibility of the received vocal
`signal has been taken into account. VAT is an
`audio conferencing tool, extending UDP [8] func-
`tionalities by providing additional features, such
`as reordering packets and filtering duplicates. We
`can report on the results obtained with respect to
`a significant subset of the encoding/decoding
`schemes used by VAT. Several very cheap boards
`are available. We tested one of them: PCM (Pulse
`Code Modulation). The other two encoding
`schemes we tested - DVI (an ADPCM encoding
`scheme by Intel) and GSM (Groupe Special
`Mobile) — use mainly software for en-/decoding.
`Compression is a particularly significant element
`in GSM, while it
`is completely absent in PCM.
`
`‘Packets generated by tested tools have almost equal size.
`
`DVI can be considered a good compromise be-
`tween bandwidth and CPU requirements. Figure
`1 presents the results obtained for these encod-
`ing schemes, with respect to the CPU time de-
`voted to encoding/decoding and the number of
`packets generated. Obviously the higher the com-
`pression factor, the higher the CPU workload. On
`the other hand the lower the compression factor,
`the higher the bit rate. As a result of our tests,
`we have proposed that
`the encoding schemes,
`with the exception of GSM, can be supported by
`systems with limited performance capabilities.
`
`% cpu PCM
`
`packets PCM
`
`% cpu DVI
`
`packets DVI
`
`ullllfl
`
`packets GSM
`
`Figure 1: Percentage of CPU time devoted to encodingl
`decoding on a Sun Sparc Station 1+ and corresponding
`number of packets generated for PCM, DVI and GSM.
`
`Carlo Gaibisso received a degree in Computer Science from Pisa University (Italy) in 1986 and joined IASI-CNR in 1989.
`His research interests are in theoretical computer science and in distributed multimedia applications. Giorgio Gambosi
`is a
`Full Professor at the Department of Mathematics, University of Rome II “Tor Vergata”. His research interests are in theoretical
`computer science and in distributed systems. Maurizio Lancia received a degree in Computer Science from University of Rome
`“La Sapienza” (Italy) in 1981 and joined IASI-CNR in 1983. His research interests are in distributed multimedia applications,
`in network management, in operating system performance evaluation and in distributed computing environments. Maurizio Vitale
`received a degree in Telecommunicationsfrom University of Rome “La Sapienza” (Italy) in 1994. His research interests are in
`distributed multimedia applications and in network management.
`
`
`
`
`
`Carlo Gaibissu, Giorgio Gambosi, Maurizio Lancia, Maurizio Virale /
`Multimedia Conferencing on Packet Switched Networks: Testing and Evaluation
`
`S 141
`
`
`
`Figure 3: Number of bytes flowing out from the access
`router interface forPCM, DVI and GSM.
`
`The limited bandwidth available has been re-
`
`vealed as the major cause of bursty traffic and
`loss of audio packets when encoding schemes
`with low compression factors are used.
`In this
`case,
`the intelligibility level
`is low.
`
`The second tool for audio conference we test-
`ed is NEVOT. The measurements we obtained
`
`with NEVOT do not differ substantially from
`those produced by VAT (and are consequently
`omitted).
`
`2.2 Video conferencing
`
`Three different tools have been tested:
`
`two of
`
`them, NV [4] and IVS [1], are public domain
`tools using software encoding/decoding. The third
`one, ShowMe,
`is marketed by SUN Microsystem
`and requires a special board for data compression.
`
`Even though some of the tools we tested in-
`tegrate audio and video conferencing, we restrict
`
`Let us now consider the transmission of audio
`
`information on a LAN. Ethernet technology has
`no difficulty supporting such transmission. In
`fact, only small numbers of collisions have been
`measured at the Ethernet interface for all encod-
`
`ing schemes, with minimal peaks when PCM is
`used. (The corresponding graphics are consequent-
`ly omitted.)
`
`The intelligibility level of the received vocal
`signal was widely acceptable, while a significant
`alteration of the vocal timbre was introduced by
`GSM. The playout delay on normal conversation
`is also acceptable. Figure 2 shows the average de-
`lay resulting from the tests.
`
`As concerns the wide area link, figure 2 shows
`the amount of traffic (in bytes per 25 seconds)
`at the interface between the access router and the
`link to the Internet?
`
`It
`
`is not difficult to see that
`
`the use of the
`
`PCM encoding scheme results in a high probabil-
`ity of bandwidth saturation, while DVI and GSM
`seem almost equivalent and do not produce a
`significant network load.
`
`
`
`
`
`Figure 2: Average playout delay measured for PCI, DVI
`and GSM.
`
`
`
`2Tests have been performed in a situation of limited network
`load. The average load in absence of audio transmission has
`been measured as about
`l0Kbps.
`
`
`
`
`
`5142
`
`Carlo Gaibisso, Giorgio Gambosi, Maurizio Lancia, Maurizio Vitale /
`Multimedia Conferencing on Packet Switched Networks: Testing and Evaluation
`
`In
`our attention to the video component here.
`fact, tests concerning audio acquisition, compres-
`sion and transmission are very close to those
`
`presented in section 2.1 and are consequently
`omitted.
`
`path through an intermediate router located at the
`National Council of Research of Rome Comput-
`ing center. These tests revealed that
`the frame
`rate dramatically decreases to less than 1.2
`frames/sec.
`
`The parameters we measured were the CPU
`time devoted to encoding/decoding and display,
`the frame rate achieved and the bandwidth re-
`
`quirements.
`
`As for audio conference tools, the last aspect
`has been evaluated in terms of the number of bits
`
`passing through the interface between the access
`router and the link to the Internet.
`
`Figures 5, 6, 7 show the amount of traffic (in
`bit/sec) at respectively the interface between the
`access router and the link to the Internet, the in-
`termediate router and the destination node.
`In
`
`each figure, both the traffic in the transmission
`direction (back of the figure) and in the opposite
`one (front of the figure) are shown.
`
`2.2.] NV
`
`7l'l00fl ’
`
`is a video conference tool based on
`NV [4]
`UDP, adopting a compression scheme based on
`DCT (Discrete Cosine Transform). The tool con-
`tinuously displays several moving icons. Each
`icon can be enlarged in images of three different
`formats. Figure [4] shows the percentage of CPU
`time (with respect to a system of medium com-
`puting power, such as a Sun Sparc Station 1+)
`spent
`in displaying a video conference session
`with respect to images of different size.
`
`100
`
`CPUVI with: ClFcolorwin:1ow
`
`CPU% wifna QI1'l'o3br wirdow
`
`Figure 4: Percentage of CPU time spent by NV.
`
`Our tests ‘proved that a display of 2 up to 7
`frames/sec can be achieved on a LAN, depend-
`
`ing on the computational power of the system
`used.
`
`As concerns the wide area link, the tests have
`
`been performed on the connection from IASI to
`the University of Rome “La Sapienza”, along a
`
`
`
`
`
`Figure 5: Number of bytes flowing at the access router
`interface.
`
`W500
`
`Figure 6: Number of bytes flowing at the intermediate
`router.
`
`
`
`Carlo Gaibisso, Giorgio Gambosi, Maurizio Lancia, Maurizio Vitale /
`Multimedia Conferencing on Packet Switched Networks: Testing and Evaluation
`
`S 143
`
`
`
`Figure 7: Number of bytes flowing at
`node.
`
`the destination
`
`It is easy to see that the bandwidth require-
`ments in all cases are almost equal to the over-
`all channel bandwidth, thus resulting in the sat-
`
`uration of the link capacity and allowing a
`frame rate of only a bit more than 1 frame/sec.
`
`2.2.2 IVS
`
`CPU?»
`
`100
`
`an
`
`613
`
`-40
`
`20
`
`Figure 9: Percentage of CPU time spent by IVS.
`
`It is also the case here that the bandwidth used
`
`is equal to the total channel capacity (the corre-
`sponding diagrams have been consequently omit-
`ted).
`
`Since IVS uses a more efficient compression
`scheme,
`this produces better results of up to 12
`frames/sec for a QCIF image on the Ethernet
`LAN and up to 2 frames/sec rate on the 64 Kbps
`Internet
`link.
`
`IVS is a video conference tool based on RTP
`
`implementing a compression scheme con-
`[5]
`forming to recommendation H.261 [9].
`
`2.2.3 ShowMe
`
`imple-
`is a Sun proprietary tool
`ShowMe [6]
`menting both video capture and data compression
`by means of a board based on a C-Cubes chip.
`
`The compression method implemented is Cell—B,
`which has been designed and developed by Sun
`Microsystem itself.
`
`Through hardware compression it is possible to
`obtain excellent frame rates, upto 25 frames/sec
`on a LAN, with a bandwith requirement
`that
`ranges from 500Kbps to 2.5 Mbps, depending on
`the characteristics of the scene. Figure 10 shows
`the results of our tests.
`
`Our tests show that, due to the greater effi-
`ciency of the compression scheme,
`the number
`of packets generated (see figure 8) is on aver-
`age substantially smaller than that generated by
`NV. Such greater efficiency is obviously paid
`for in terms of computing resources (figure 9).
`
`Packets
`
`2 5 U
`
`2130
`
`150
`
`100
`
`5|] ‘
`
`Figure 8: Number of Packets generated by IVS.
`
`
`
`
`
`S144
`
`Carlo Gaibisso, Giorgio Gambosi, Maurizio Laricia, Maurizio Vitale/
`Multimedia Conferencing on Packet Switched Networks.‘ Testing and Evaluation
`
`
`
`
`
`Bandwidth for
`
`high moving scene
`(kbit/sec)
`
`
`
`Bandwidth for
`
`low moving scene
`(kbit/ sec)
`
`1400-1600
`
`2300-2700
`
`Bandwidth for
`
`still scene
`
`
`
`
`(kbit/ sec)
`
`
`
`Frame rate
`
`
`(frame / sec) Characteristics
`
`
`
`
`Ix)U1
`
`IQU1
`
`384x288 pixel
`128 colors
`
`384x288 pixel
`128 colors
`
`176x144 pixel
`32 colors
`
`176x144 pixel
`32 colors
`
`384x288 pixel
`128 grayscale
`
`384x288 pixel
`128 grayscale
`
`176x144 pixel
`32 grayscale
`
`176x144 pixel
`32 grayscale
`
`
`
`
`
`300-400
`
`1100-1300
`
`350-450
`
`700-800
`
`90-140
`
`230-300
`
`1000-1200
`
`1800-2000
`
`270-300
`
`1000-1 1 00
`
`300-400
`
`600-700
`
`60-100
`
`200-210
`
`Figure 10: ShowMe compression performances.
`
`On a WAN ShowMe results in worse perform-
`ance,
`in terms of frame rate,
`than NV and IVS:
`
`the compression schemes the latter use
`in fact,
`present a better compression factor than Cell-B.
`
`3. An alternative approach: MPEG
`
`The experimental results reported in section 2
`clearly show that narrow bandwidth is the major
`restriction to video conferencing on a WAN, thus
`introducing the opportunity of evaluating the im-
`pact of encoding methods with a higher compres-
`sion factor.
`In fact, as long as the computing
`power is sufficient to handle the required encod-
`ing/decoding operations,
`the transmission rate is
`determined by the compression factor.
`
`We concentrated our attention on the MPEG
`
`[10] compression scheme. MPEG encoding gen-
`erates three different types of frames:
`
`- an intra—frame encoding based on DCT,
`which reduces spatial redundancies inside the
`single frame (I—frame);
`
`- an inter—frame encoding based on motion
`pre—diction (P—frame);
`
`- an inter—frame bidirectional motion predictive
`encoding(B-frame).
`
`The achievable compression factor depends on
`the particular pattern following which I, B and
`P—frames appears in the encoding sequence.
`
`MPEG adopts moving compensation techniques
`similar to those employed by H.261, from which
`it
`is mainly distinguished by a bidirectional
`motion prediction. This feature guarantees higher
`compression factor and makes backward/forward
`searching possible and gives direct access to the
`inside of the video sequence.
`
`
`
`
`
`Carlo Gaibisro, Giorgio Gambosi, Maurizio Lancia, Maurizio Vitale /
`Multimedia Coriferencing rm Packet Switched Networks.‘ Testing and Evaluation
`
`S 145
`
`In this section the MPEG encoder and decod-
`
`er performances are presented, with the main ob-
`jective of proving that such software could actu-
`ally support teleconferencing applications.
`
`The efficiency of the en/decoding phases is
`highly dependent on the image size,
`the pattern
`of I, B and P—frames and the characteristics of the
`scene.
`
`We are mainly interested in obtaining accept-
`able values of the frame rate in playout. It is well
`known that,
`the frame rate achievable by the
`decoder‘ is definitely higher than that achievable
`by the encoder. As a consequence we restricted
`our attention to the encoder performances.
`
`We based our test on 10 YUV (luminance and
`
`chrominance — 320x240 pixels) frames sample.
`Figure 11 shows the performances in terms of
`compression factor, frame rate and bit rate for
`two different hardware platforms.
`
`Clearly, systems with low computing power
`can not effectively support the coding activity,
`while systems with higher computing resources,
`such as Alpha DEC AXP 3000/500, guarantee ac-
`ceptable frame rates, which improve considerably,
`up to 5-7 frames/sec, when smaller images are
`processed. As shown in figure 11, acceptable
`frame rates via software can be achieved only at
`the expense of the compression factor,
`thus not
`fully exploiting the potentialities of MPEG. On
`the other hand, hardware solutions are currently
`available but
`their cost
`is exorbitantly high.
`
`A distributed version of the encoder, developed
`the University of California at Berkeley, has
`at
`also been tested on 4 Sun Sparc Station 1+,
`achieving a 2.5-3 frames/sec rate.
`
`We have not yet extensively tested the per-
`formance of MPEG when transmitting on LANS
`and WANs.
`
` .
`E Compression
`
`Theoric Bit rate
`for 30 frame/ s
`
`Hardware
`
`PC 386
`
`888188
`I 818888
`I 888188
`888828
`DEC AXP I 1-828571
`I 888888
`I 888188
`B=B-frame search
`P=P-frame search
`
`p 1=cross2
`p 2=simple
`
`b=logarithmic
`
`Figure 11: Coder performances for two different hardware platforms and 352x240 pixels images.
`
`3 486/DX2 systems guarantee frame rates ranging from 4 to
`21 frames/sec depending on the image size.
`
`
`
`
`
`S146
`
`Carlo Gaibisso, Giorgio Gambasi, Maurizio Lancia. Maurizio Vitale /
`Multimedia Conferencing on Packet Switched Networks.‘ Terting and Evaluation
`
`On a LAN it seems that the frame rate achiev-
`
`able by the encoder and the number of frames per
`second received at the destination node are quite
`similar. The situation seems to be completely dif-
`ferent if data is transmitted on a WAN. In this
`case in fact the frame rate at the receiver is sub-
`
`stantially lower then that at the transmitter. It is
`our opinion that this is caused by the overhead
`introduced by TCP transport protocol.
`
`It is worth noting that use of a connectionless
`transport protocol (i.e. TCP) is recommended in
`this context. In fact, since the B, P and I-frames
`
`in a MPEG coded sequence are strictly correlat-
`ed,
`losing one frame could make a lot of other
`frames undecodable.
`
`4. Conclusion and Future Work
`
`As a result of the tests performed,
`sible to state the following:
`
`it
`
`is pos-
`
`1.Audio conferencing does not seem to repre-
`sent a real problem for either LANs or
`WANs. The transport protocols and
`compression schemes we considered seem
`sufficient to obtain good performance.
`
`2.With regard to video conferencing, it seems
`that
`the compression factor of the scheme
`used is crucial in obtaining efficiency. This
`is true as long as we assume that a relevant
`fraction of computing power can be devoted
`to activities related to video transmission
`
`(mainly encoding/decoding).
`
`3.As a consequence, MPEG represents a good
`solution for achieving a relatively high frame
`transfer rate: the main bottleneck seems to be
`
`the inefficient encoding process typical of
`MPEG, which nevertheless allows frame
`rates on the order of 1.5-2 frames/sec.
`
`4.Another problem with MPEG is determined
`by the strict correlation between the
`encoding of successive frames: packet loss
`may result in difficulties reconstructing
`
`images in sequence. This produces the need
`for using a more secure transport protocol
`such as TCP instead of the more efficient
`
`UDP. Such a problem will probably be
`solved with the availability of new versions
`of MPEG encoders/decoders.
`
`Future work along this line should concentrate
`
`on a more in—depth evaluation of the applicabil-
`ity of MPEG to video conferencing.
`In particu-
`lar, we plan to test the efficiency determined by
`using hardware devices (especially for encoding)
`or parallel/distributed encoding/decoding algo-
`rithms (for example implemented in the DCE [11]
`environment).
`
`References
`
`[1]
`
`[2]
`
`[3]
`
`[4]
`
`[5]
`
`[6]
`
`[7]
`
`[8]
`
`[9]
`
`T. Turletti, IVS, “H.261 Software Codec for Videocon-
`ferencing Over the Internet”, Research report No 1834,
`INRIA ,January 1993.
`
`S. Deering, Mbone, “Host Extensions for [P Multicast-
`ing", RFC 1054.
`
`H. Schulzrinne, NEVOT, “Voice Communication Across
`the Internet: A Network Voice Terminal", 1993.
`
`R. Fredericks, NV, “NV network videotool”, available
`for anonymous ftp from ftp.parc.xerox.com, 1993.
`
`H. Schulzrinne, S. Casner, RTP, “RTP: ATransport
`Protocol for Real-Time Applications”,
`INTE