Database Reference
In-Depth Information
900
800
700
600
500
400
300
200
100
0
10
20
30
40
50
60
70
80
90
100
110
time (day)
Fig. 8.
Consecutive day using measure d 2 .
appearing as background noise. Figure 8 shows the results for distance mea-
sure
d 2 being applied to consecutive graphs in the time series. Whilst this
measure considers both topology and trac, it does not appear to provide
any additional indicators of change additional to those found using the
topology only measure. The main difference with this measure is that it
has increased the amplitude of the first two peaks so that they now appear
as major peaks. This suggests that these peaks were a result of network
change consisting of large change in edge weights.
The findings for this technique indicate that the application of measures
to consecutive graphs in a time series is most suited for detecting daily
fluctuations in network behaviour. This is especially useful for identifying
outliers in a time series of graphs. Ideally, it would be desirable that each
point of significant change be validated against known performance anoma-
lies. Unfortunately, information generated by the network management
system of the network under observation is not sucient for validation
purposes, because all data used in the experiments were collected from a
network under real operating conditions. That is, the data are unlabelled
as there is no 'super-instance' that would know whether a network change
Search WWH ::




Custom Search