beautypg.com

Measure end-to-end packet loss, Adjust ping measurements – Nortel Networks NN43001-563 User Manual

Page 146

background image

146

ITG engineering guidelines

Measure end-to-end packet loss

The PING program also reports if the ICMP packet made its round trip
correctly or not. Use the same PING host setup to measure end-to-end
error. Use the same packet size parameter.

Sampling error rate, however, requires taking multiple PING samples,
at least 30 to be statistically significant. Therefore, obtaining an error
distribution requires running PING over a greater period of time. The error
rate statistic collected by multiple PING samples is called Packet Loss
Rate (PLR).

Adjust PING measurements

Make adjustments to the PING statistics as required in the following
situations.

One-way as compared to roundtrip

The PING statistics are based on round trip measurements, where the QoS
metrics in the Transmission Rating model are one-way. In order to make
the comparison compatible, the delay and packet error PING statistics are
to be halved.

Adjustment caused by IP Trunk 3.01 (and later) processing

The PING measurements are taken from PING host to PING host. The
Transmission Rating QoS metrics are from end-user to end-user and include
components outside the intranet. The PING statistic for delay must be
further modified by adding 93 ms to account for the processing and jitter
buffer delay of the IP Trunk 3.01 (and later) nodes. No adjustment has
to be made for error rates.

If the intranet measurement barely meets the round trip QoS objectives,
there is a possibility that the one-way QoS is not met in one of the direction
of flow. This can be true even if the flow is on a symmetric route, due to the
asymmetric behavior of data processing services.

Late packets

Packets that arrived outside of the window allowed by the jitter buffer are
discarded by IP Trunk 3.01 (and later). To determine which PING samples to
ignore, first calculate the average one-way delay based on all the samples.
Add 500 ms to the average. This is the maximum delay. All samples whose
one-way delay exceeds this maximum are considered late packets and
removed from the sample. Calculate the percentage of late packets and
add that to the packet loss statistic.

Nortel Communication Server 1000

IP Trunk Fundamentals

NN43001-563

01.01

Standard

Release 5.0

30 May 2007

Copyright © 2007, Nortel Networks

.