Вы находитесь на странице: 1из 9

MakingSenseofBroadbandPerformance

NeverMindtheConnectionSpeed, MeasuretheConnectionQuality!

IntodaysInternetworlddominatedbyondemandVoice, VideoandIPTV,thekeytogoodconnectionperformanceis consistentthroughputnotfastspeeds.

Nevermindtheconnectionspeed,measurethequality!

Nevermindtheconnectionspeed,measurethequality!

KnowingyourInternetspeedisslowwithoutprovidingthemeanstohelp resolvetheproblemisoflittlevalue.
WhendiscussingInternetconnectivityproblemsthenumberonecomplaintisalwaysaboutspeedorthe lackofit.Endusersrelaythesamemessageoverandoveragain,andquitefrequentlyuserswith InternetserviceproblemsarequicktodefendtheirchoiceofserviceproviderwithstatementslikeButI havea5Mbpsora10MbpsconnectionsoIshouldnthavejerkyvideoorButthespeedtesterIuse saysIcanrun9VoIPlines,sowhyaremyVoIPcallsoftengarbled? AsspeedistheprimecriteriawhenusersselectanInternetserviceprovider,whentheyareaskedwhat isthequalityoftheservice?,thequestionismostoftenmetwithapuzzledalongwiththeretort, Quality!Whatdoyoumeanbyquality?

SpeedversusQualityisacrucialissueformeasuringbandwidthperformance.Anexampleof speedwithoutqualitywouldbelikedrivingaFerrariduringtheheavilycongestedrushhour whilethecapabilityofhighspeedexists,thespeedisnotrealizedduetootherfactors.


Wealldriveonhighwaysandthroughexperienceweknowonlytoowellthatourjourneytimewillnot begovernedbythemaximumspeedofourcarorthemaximumspeedofthehighway.Inrealityitwill begovernedbythemanydifferenteventsthatoccuronourjourneysuchasweather,congestion, accidentbackuporhighwayconstruction.

Runningaspeedtestthatshowsaslowthroughputwithoutconsiderationoftheevent informationtohelptheISPtoidentifyandresolveinherentperformanceissueshasabsolutely novalue.Onlythroughcarefulmeasurementofeveryaspectoftheendtoendjourneycana speedtestreallystarttobenefittheuser,andmoreimportantlybenefittheserviceproviderby providingtheinformationneededtoidentifytheproblem.


IntheInternetworldtherearemanydifferentissuesthataffecthowourapplicationsperform.When performanceisasweexpectwearecontent,whenitisnotwegetfrustrated.Understandingtheevents thatimpactyourInternetconnectionalongwiththequalityofservicedeliveredforyourapplicationsis vitallyimportantfortheproblemstoberesolved.Inourhighwayexample,ifthequalityofthehighway servicewastobemeasuredbyitsusers,itwouldbeameasureofthehighwaysabilitytodeliveryouto yourdestinationinatimethatcloselymatchesthetimeitwouldtakeifyouwereabletosustainthe

2007VisualwareInc.

Page1

Nevermindtheconnectionspeed,measurethequality!
maximumallowedspeed.Inotherwords,assumingyouaredrivingasfastasallowedbythelegallimit, thecalculationofhighwayqualitywouldbe: HighwayQuality%=YourSpeedx100 maximumallowedhighwayspeed

Ifthehighwayislegallylimitedto65mphandyouraveragespeedis35mphthenthequalityis: HighwayQuality%=35x100=53% 65

Thequalitymeasureiscriticalbecauseifthehighwayqualityissuchthatyouareonlyabletomaintain 30mphwhentheapplication(arrivingattheairporttocatchtheflight)requires35mph,youwill certainlymissyourflight.

Iknowhowtomeasurespeed,howdoImeasurethequality?
ThekeytosuccessofanInternetconnectionisacombinationofagoodspeedwithagoodqualityof service.Infact,itispreferabletohaveaslower3Mbps(Megabitspersecond)connectionwitha99% qualityofserviceratherthana6Mbpsconnectionwitha50%qualityofservice.Bothwillachieveabout thesamethroughputoverall,howeverthedelaysinherentinthepacketflowthatresultinthelowerof qualityofservicewilladverselyimpacttimedependantapplicationssuchasVoIP,videoorMP3. Inordertomeasurequality,anInternetconnectiontestingapplicationhastomeasuretheeventsfor theInternetconnectionaswellasthespeed.Otherwise,youonlyknowthatthetrafficisslowbutyou havenoclueastothecause.Tobetterascertainthecauserequiresthatthetestingapplicationis capableofmeasuringthetrafficflowaswellasthetrafficspeed.Graphicalviewsofthetrafficflowshow theinherentdelaysoccurringwiththedataalongwiththeallimportanttiming. Oncethedelaypictureisexposed,youcanstarttoseetheimpactthedelayshaveonthedata movementandthuslyontheapplications.Forexample,inlinewithouranalogyofdrivingtotheairport, youwillbeabletoseethosepartsofthedatapacketjourneythatarefastandthosepartsthatarenot. Youwillalsobeabletoseeifthefastdatamovementperiodsarereachingthecontractedmaximumor whether,asisoftenthecase,thefastperiodsarestillnotmeetingthecontractedservicelevels.In addition,thevolumeofdelays,sizeofthedelaysandthetimingofthedelaysprovidetheinsightto understandwhethertheproblemistrafficrelated(i.e.Internetcongestion),orwhetheritispolice related,i.e.someothersortofdatatrafficmanagementortimemultiplexingisbeingimposedbythe ISP.Thechartbelowshowsagoodexampleofspeedversusquality.

2007VisualwareInc.

Page2

Nevermindtheconnectionspeed,measurethequality!

Fig.1MySpeedchartshowingdatatransferdelays

InthechartplottedabovetheTCPdelay(redline)suddenlyimpactsthedataspeedtestatabout4 secondsintothetest.Thethroughputorbandwidthspeed(blueline)dropsfrom3Mbpsto approximately0.2Mbpsbeforerecoveringandcompensatingsome2secondslaterwiththroughputat3 Mbpsandspikesto5,6andeven7Mbps.Aconventionalspeedtesterwouldmostprobablyhave indicatedanaveragespeedofaround3.5Mbps.Themeasurementwouldmostprobablybeacceptedas OKbythetesterasthespikeselevatethetesttobecloseenoughtothecontractedserviceof4Mbps.In reality,thereareanumberofissuesthatneedtobeaddressedbecausethesuddendropwassevere andalsosustained(2seconds).Notethatwithpeakbandwidthshowingat5Mbpsto7Mbpsthe connectionisabletocompensateforthedropwithhigherdataspeedstobringtheaverageupto3.5 Mbpsormore.Unfortunatelytheaveragespeedisfinebutthequalityispoor.Itislikelythatthequality ofaVoIPcallorvideosessionwouldsufferundersucherraticandunpredictableconditionswithTCP delaysexceeding180ms(milliseconds)ormore.

2007VisualwareInc.

Page3

Nevermindtheconnectionspeed,measurethequality!

Fig.2MySpeedchartshowingdatatransferdelays

Thelongerthetimedelaybetweendatatransfers,themoredramatictheimpactonservicequality.In Fig.2itwouldhavebeenverydifficulttomaintainanyqualityofserviceasthedelaywasalmostafull3 secondsandthedatathroughputdroppedtobelowmodemspeedsfortheduration. Whatshouldagoodqualityconnectionlooklike?

2007VisualwareInc.

Page4

Nevermindtheconnectionspeed,measurethequality!

Fig.3MySpeedchartshowinggoodconnectionquality

Interestinglyenough,thetwocharts(Figures2and3)aboveweretakenfromthesameofficelocation butfromtwodifferentPCsconnectedtodifferentISPs.Thedistancetothetestingserverwasidentical, justtheISPsweredifferent.Thefirstchartwasa4Mbpsconnectionandthesecondwasa1.5Mbps connection.Inthecaseofthelatter(Fig.3)youcanseetheserviceisverycleanandreflectsaconstant andconsistentdataflow.Thedelaysareaconsistentlylow7msandeventhepeakdelaysareonly15 ms. Understandingthenatureofthedelayssuchastiming,sizeandfrequencyprovidesagoodclueastothe cause.Thetwoprimecausesaretrafficcongestionandtrafficcontrol.Congestionisaverycommon problembecauseISPsoftenoversubscribecircuitsinthelocalexchanges.Controlisalsocommonas ISPsdividehighbandwidthconnectionsusingtimesharingalgorithms.Inotherwords,youthinkyou havea5Mbpsconnectionwheninrealityyouhavearegulated5Mbpssliceofa100Mbpsconnection. Thisisacommonapproachofcableproviders.

2007VisualwareInc.

Page5

Nevermindtheconnectionspeed,measurethequality!

Fig.4MySpeedchartshowingregulatedthroughput

Ifthedelaysaremanmade,orbydesignratherthanrandomchance,thenaregularpatternwillquickly show.InFig.4abovethereisa200msdelayevery2seconds.Youcanseetheimpactofthatdelayon thedataflowandwithittheimpactonapplications,especiallytimesensitiveapplicationssuchasVoIP orvideo. Intermsofthrottlebasedbandwidthmanagementasshownabove,toalargeextenttheimpactwill dependontheintelligenceandsophisticationoftheimplementation.Forexample,manysmalldelays versusfewerlargedelayswilldeliverdifferentcharacteristicsinthedataflow.

2007VisualwareInc.

Page6

Nevermindtheconnectionspeed,measurethequality!

Fig.5MySpeedchartshowingpoorregulatedthroughput

Fig.5aboveshowsamoredramaticimpacttothedataflowoverFig.4asspeedsdegradetobelow 200k.Noticethatinmostcasesthedataflowalsorecordsahigherthannormalspikeafterthedelay, whichimprovestheaveragespeedperformance,butnotnecessarilytheperformancequality.Thisoften leadstheusertobelievethatthespeedisacceptable,butthencannotunderstandwhyVoIPcallskeep dropping. InextremecasesstaggeringTCPdelaysarebeingrecorded,andtheuserstillbelievesthatthespeed averageisacceptable.

2007VisualwareInc.

Page7

Nevermindtheconnectionspeed,measurethequality!

Fig.6MySpeedchartshowingdramaticdelayimpact

Fig.6aboveshowssuchanextremecase.Duringthefirsthalfifthetest(about4seconds)theaverage throughputrecordedjustexceededthecontracted1.5Mbps.Inthelasthalfofthetestitwaslessthan 0.5Mbps.Isthatagood1.5Mbpsthroughput?Inrealitytheconnectionshows10Mbpspeakswith0.0 Mbps(zero)troughs.Thisisnotaqualityconnectionbyanyaccountandisincapableofsustainingtime dependentapplications. Inshort,knowingthespeedofaconnectionisonlyasmallpartofthepicture,anddoesnothingto identifyaconnectionperformanceproblem.Thelargerpartofthepictureistheconsistencyofthedata flowconnectionswithwidevariationsinspeedsorlargegapsbetweendatatransferswillcause applicationslikeVoIPandvideotoperformpoorly.Totrulyunderstandconnectionperformance,good orbad,itisessentialtomeasurethequality.

Webapplicationsneedaqualityconnectionnevermindthespeedmeasurethe quality!
ForinformationonhowMySpeedcanhelpyoumeasure connectionquality,pleaseseewww.myspeed.com.

VisualwareInc.
POBox668 Turlock,CA95381 8668479273 www.visualware.com

2007VisualwareInc.

Page8

Вам также может понравиться