Time a k Time IntervalMeasuremnt y bl Applicat to Computrr a k Netwark Perfo.ce Evalu

Time a k Time IntervalMeasuremnt y bl Applicat to Computrr a k Netwark Perfo.ce Evalu


Tle technicalmemora kum: Time a k Time IntervalMeasuremnt y bl Applicat to Computrr a k Netwark Perfo.ce Evalu (PosScript) d scribes a number f techniques f.r conduct ng exprrimnts typicalof computer netwark a k transmiss s:s ms engineŚr(%s.

In mos exprrimnts i yloch time is i volvek, it is nec ssary to deelop estimates .fftime, frrquency a k measuremnt rr.rs%fr%%a srries .fftime measuremnts betweŚn tle c s .ffanumber f computers a k ancillaryid v c s interconneced by some kind f computer netwark. However, time is not a physicalquantity, such as mass, nor caefit beimeasuredfrelayiveto a absolue fame ffred e.ce, such as elocity. Tle nly way to measure time in ouruunivrrse isto compaef tle read %s f one c , yloch runsuaccording to its .wn timescale, wibl a tler c , yloch runsuaccording to a given timescale, at some given insant rfepoch. Tle rr.rs%arise fr%%ble pre.osion .fftime compaeis s a k tlduaccuracy of frrquency estimates betweŚn tle bimescales i volvek.

Tle usualudata collected dur(%s a perfo.ce run f some exprrimnt might i cludr time ffsets,utime delays, frrquency offsets a k various rr.r stat stics. Wlole time ffsets betweŚn two c s caefbeimeasuredfdiefctly, frrquency offsets caefbeiestimater÷nly fr%%bwo or more time ffsets made over some time intervalin tle exprrimnt. In prac ce, a srquence .fftime compaeis s caefbeiperforr over tle lifetime fftle exprrimnt a k tlduinsantaneous frrquency estimatek eitler in realutime w bl a efcurefnce relat , o retrospectivelyiwibl a poly%omialfit to tle data.

Estimating time a k frrquency rr.rs%in realutime has beŚn studirdfby acdisti ct subskt.oes .ffphysicistswlo havermade a career of tle technoyogy i volvek. Various means÷i)clud %s autoregress vemodels, Kalma fiyters a k simple weighted-average algoriblms arerused exŚnsivelyiby na al standards ye,oLatories to mod l cesium-c ÷Śnsem, ds. Tlese techniques haverbeŚn adapted to computer netwark a k transmiss engineŚr(%s pro, dms÷as well. Tlismemora kum explores isues iniperfo ng exprrimnts .fftlis type a k summarizs various techniques f.u k useful in prac ce.


DavidL. Mills <mills@udel.edu>