Side_1_360

3 Measurement Methods and

Infrastructures

3.1 Introduction

The specification of an actual measurement and
monitoring platform depends on what perfor-
mance parameters that are to be observed. The
measurement platform is characterized by the
following properties:

A) Measurement points – at which points in the
network it is measured (end user, interface
card in end user equipment, routers, switches,
servers, access network, edge routers, etc.).

B) Measurement method – active (intrusive) or
passive (non-intrusive).

C) Handling and post-processing of measure-
ments – data reduction techniques (accumula-
tion of statistics, selection of packets or
flows), accuracy in measurements (observa-
tion period, correlation, other).

D) Measurement period – time interval over
which the measurements are collected (time
of day, continuously or by sampling, etc.).

Systematic gathering of measurement data from
a communication infrastructure requires the
establishment and deployment of a measurement
infrastructure. The measurement infrastructure,
as illustrated in Figure 3.1, consists of a set of
measurement units3)placed at strategic locations
(measurement points) in the network domain.
The measurement method dictates the actual
implementation of the measurement infrastruc-
ture. Today several projects develop measure-
ment infrastructures to collect measurement data
from the Internet, examples of such efforts are
[NAI], [McGregor], [NIMI], [RIPE] and [Sur-
veyor].

The two principal methods to collect network-
level measurements are; either actively by insert-
ing probe packets [RFC792] [AMP] [RIPE]
[Surveyor] or passively by observing real pack-
ets [Claffy97] [Brownlee] [Cflowd] [Net-
Flow99] [Careces91] [Claffy98] [Viken99].
Active and passive measurements are discussed
further in Section 3.3 and Section 3.4, respec-
tively.

Collected measurement data is subject to post-
processing and analyses. There are several
options regarding the post-processing and analy-
ses of measurement data, as to which data reduc-

tion techniques to use and where the post-pro-

cessing units (local vs. central processing of

data) should be located. The various approaches

have different resource needs and offer different

measurement accuracy.

3.2 Timestamp Requirements

Active measurements add a timestamp to every

probe packet sent while passive measurements

associate a timestamp with the observation of a

packet. Thus, obtaining accurate timing informa-

tion is crucial to minimize the measurement

error for both active and passive delay measure-

ments. Accurate estimation of one-way delay

requires the clocks of the measurement units to

be synchronised, accurate and have high preci-

sion. These issues are discussed in e.g.

[RFC2330] [Pasztor].

The clocks of the distributed PCs collecting the

measurement data can be synchronised by using

GPS receivers [Surveyor] [RIPE] or the network

time protocol (NTP) [RFC1305]. NTP can only

provide timestamp accuracy in the range of mil-

liseconds while the accuracy of the output signal

from modern GPS receivers is well below one

microsecond.

Further, generating timestamps in software intro-

duces an additional measurement error due to

operating system scheduling. Different operating

systems give different measurement errors

[Pasztor]. To achieve very accurate timestamps,

the measurement units must be synchronized by

GPS receivers and the timestamps have to be

Figure 3.1 Example of a

measurement infrastructure

3)A measurement unit is not necessarily a dedicated physical unit. The measurement functionality

can be integrated in the hardware or software of a node.

Monitor

Network domain

Measurement

point