190
Measuring Intranet QoS
Late packets
Packets that arrive outside of the window allowed by the jitter buffer are discarded. To determine
which
samples to ignore, calculate the average one-way delay based on all the samples. Add
Ping
300 ms to that amount. This amount is the maximum delay. All samples that exceed this one-way
delay maximum are considered late and are removed from the sample. Calculate the percentage of
late packets, and add that percentage to the packet loss statistics.
Measurement procedure
The following procedure is an example of how to get delay and error statistics for a specific site
pair during peak hours.
Program a script to run the Ping program during the intranet peak hours, repeatedly sending a
series of 50 Ping requests. Each Ping request generates a summary of packet loss, with a
granularity of 2%, and, for each successful probe that made its round-trip, that many rtt samples.
For a strong network there must be at least 3000 delay samples and 60 packet loss samples. Store
the raw output of the
one-way delay and packet loss.
Repeat this for each site pair. At the end of the measurements, the results are as shown in the
following table.
Table 33 Delay and error statistics
Destination
pair
Santa Clara/
Richardson
Santa Clara/
Ottawa
Santa Clara/Tokyo
Richardson/Ottawa
Richardson/Tokyo
Ottawa/Tokyo
Other measurement considerations
The Ping statistics described above measure the intranet before IP telephony installation. The
measurement does not take into consideration the expected load provided by the IP telephony
users.
P0606298 02
results in a file. Determine the average and standard deviation of
Ping
Measured one-way delay
(ms)
Mean
Mean+
171
179
Measured packet loss (%) Expected QoS level
Mean
Mean+
2
2.3
Mean
Mean+
Good
Good