Digital Signal Processing Reference
In-Depth Information
x 10
5
(a)
(c)
(e)
2.5
−
1
−
1
−
1.5
2
−
2
−
2
1.5
−
2.5
−
3
−
3
1
−
3.5
−
4
0.5
−
4
0
−
5
−
4.5
0
5
10
x 10
4
0
2
4
6
0
1
2
3
Time
log10(x)
log10(lag)
x 10
5
(b)
(d)
(f)
2.5
−
1
−
1
2
−
2
−
2
1.5
−
3
−
3
1
−
4
−
4
0.5
5
5
0
−
−
0
5
10
x 10
4
0
2
4
6
0
1
2
3
Time
log10(x)
log10(lag)
FIGURE 6.5
The actual single-user network traffic (first row), the synthesized traffic traces (second row), and
their corresponding LLCD plots and codifference estimates (only one synthesized trace is shown
in b). (From X. Yang and A.P. Petropulu,
IEEE Trans. Signal Process.,
49(7), 1349-1363, July 2001.
c
2001 IEEE. With permission.)
Because EAFRP is constructed for modeling single-user traffic, aggregated
traffic modeling can then be realized by superimposing single EAFRPs. In-
deed, it was shown in Reference 20 that the superposition of EAFRPs is
marginally heavy-tail distributed and long-range dependent in the gener-
alized sense. Therefore, the EAFRPs and their superpositions can capture the
two most relevant statistical characteristics of high-speed network traffic, i.e.,
impulsiveness and self-similarity.
In Reference 34, a further modification of the EAFRP model was proposed.
This was motivated by the distinctive two-slope appearance of the log-log
complementary distribution of real traffic data (see Figure 6.3), and the na-
ture of the true bounds on the user transmission rates in real networks was
proposed. Limits on the sender's and the receiver's TCP (Transport Control
Protocol) window sizes, TCP congestion avoidance strategies, and bandwidth
bottlenecks within the end systems are among the many reasons that lead to
Search WWH ::
Custom Search