Side_1_360

• Class 1 – throw away any unwanted informa-
  tion contained in the packet trace. The trace
  can be filtered such that only selected packets
  are kept, e.g. packets sent between certain
  source and destination pairs. Further, any field
  in the packet header that is not interesting in a
  given context can be thrown away.


• Class 2 – statistical reduction, e.g. computing
  the average and variance for selected metrics.

The disadvantage of any data reduction is that
single item information in the original measure-
ment data is lost and cannot be reconstructed
from the processed measurement data.

The post-processing of measurement data, as
shown in Figure 3.5, can be performed locally at
the measurement unit (local data reduction) or at
a central host (central data reduction). Further,
the local host can perform the post-processing
“off-line” or “real-time”. Thus, the following
strategies to process the measurement data must
be considered:

Central Processing of Measurement Data
The measurement units collect and store raw
packet traces captured during a given measure-
ment period. Then the raw data is exported to the
central host where post-processing of the mea-
surement data is performed.

Local “off-line” Processing of Measurement
Data
The measurement units collect and store raw
packet traces to permanent storage locally. After
each measurement period, the raw measurement
data is post-processed and only the processed
data is exported to the central site.

Local “real-time” Processing of Measurement
Data
The measurement units extract relevant informa-
tion from every packet and perform the post-pro-
cessing in real-time without writing raw data to
permanent storage. Thus, only processed data is
stored locally and exported to the central host.

Some selected examples of post-processing of
raw packet traces are presented to illustrate the
resources required by the various approaches.
The following assumptions are made for the
examples:

• nmeasurement units capture packet traces at
  selected measurement points. Each unit cap-
  tures traces from two links (e.g. each direction
  of a bi-directional link) with capacity, c[Mb/s],
  at full line rate. The links carry traffic with an
  average packet length, m[Bytes]. Thus, the
  arrival rate of packets to each unit equals
  . Let n= 20, = 200 [Bytes] and

c= 155 [Mb/s].

• The measurement period, t, considered has a
  duration of 60 minutes.

The following methods for grouping, sorting and

data reduction techniques of raw packet traces

are considered:

Method A) Unidirectional performance

without data reduction

All available information about packets, in-

cluding both endogen and exogen attributes,

is needed. Assume that bbytes are required

to store all information about a single packet.

Letbbe equal to 64 bytes.

Method B) Unidirectional performance

• filtering of packets
  All available information about every packet
  satisfying certain requirements are kept (e.g.
  type of service equal to a certain value).
  Assume that ten percent of the packets meet
  the requirements.

Method C) Flow records

All information about every flow is collected.

Assume that on the average a flow consists of

fpackets and that bbytes are required to store

all information about a flow. Let fbe equal to

40 packets per flow. It may be noted that the

number of packets per flow depends on how

a flow is defined.

Method D) Single point metrics

Statistical data reduction of the measurement

data is performed. To determine the average

and variance, and are computed

for certain metrics (e.g. throughput, packet

length, interarrival time etc.). Let mdenote

number of variables computed while kis the

number of bytes required to store each of the

variables. Let mand kbe equal to 100 metrics

and ten bytes, respectively.

Table 3.3 shows the data volumes stored locally

at each monitor, totally exported and stored cen-

trally by the various methods described above.

Note that which approach to apply depend on the

performance metrics to be observed.

∑li ∑li^2

λi= c m

m⋅ 8

Figure 3.5 Post-processing of

measurement data

Sorted/

grouped

data

Raw data

Sorting/

grouping

Post-

processed

data

Data

reduction