146 ITG engineering guidelinesNetwork delay and packet loss evaluation exampleFrom PING data, calculate the average one-way delay (halved from PINGoutput and adding 93 ms IP Trunk 3.01 (and later) processing delay) andstandard deviation for latency. Do a similar calculation for packet losswithout adjustment.Adding a standard deviation to the mean of both delay and loss is forplanning purposes. A customer might want to know whether trafficfluctuation in their intranet reduces the user’s QoS.Table 28 "Sample measurement results for G.729A codec" (page146) provides a sample measurement of network delay and packet loss forthe G.729A codec between various nodes.Table 28Sample measurement results for G.729A codecMeasured one-waydelay (ms)MeasuredPacket loss(%)Expected QoS level(See Table 33 "IP Trunk3.01 (and later) QoSlevels" (page 149))Destinationpair Mean Mean+σ Mean Mean+σ Mean Mean+σSanta Clara/Richardson171 179 1.5 2.1 Excellent GoodSanta Clara/Ottawa120 132 1.3 1.6 Excellent ExcellentSanta Clara/Tokyo190 210 2.1 2.3 Good GoodRichardson/Ottawa220 235 2.4 2.7 Good GoodAs an example, the delay and loss pair of traffic from Santa Clara toRichardson (171 ms and 1.5%) will meet "excellent" criterion, but theircounter part with standard deviation (179 ms and 2.1%) can achieve only"good" QoS.Since the algorithm implemented in IP Trunk 3.01 (and later) calculates onlymean and not standard deviation, it confirms the "excellent" rating (if theobjective is set for excellent, it will not fallback to alternate facilities), but thecustomer has up to a 50% chance of experiencing a service level inferior toan "excellent" level.Nortel Communication Server 1000IP Trunk FundamentalsNN43001-563 02.01 StandardRelease 5.5 21 December 2007Copyright © 2007, Nortel Networks.
