Simulation and Wavelet Analysis of Packet Traffic
Bruce Chen, Zelimir Lucic, and Ljiljana Trajkovic, {bchenb, zlucic, ljilja}@sfu.ca, http://www.ensc.sfu.ca/research/cnl Communication Networks Laboratory, School of Engineering Science, Simon Fraser University
1. Traffic:
• Complex traffic patterns arise from multiplexed data, voice, and video • Traditional traffic models fail to capture essential traffic characteristics • Traffic often exhibits long-range dependent (self-similar, fractal) behaviour • Current traffic models should capture long-range dependent traffic characteristics
2. Simulation topology and scenario:
• We analyze the impact of traffic on the Quality of Service (QoS) in packet networks • We use trace-driven network simulations (using ns-2) • MPEG-1 traffic is transmitted over UDP/IP (User Datagram Protocol/Internet Protocol) • UDP is suitable for real-time applications because of small delay • Buffer size of the router is set according to delay requirements • Router employs five different queuing schemes:
Terminator 2 MPEG-1 trace
Trace Silence of the Lambs Terminator 2 MTV Simpsons Talk Show 1 Jurassic Park 1 Mr. Bean News Star Wars Talk Show 2
Mean bit rate (Mbps) 0.18 0.27 0.49 0.46 0.36 0.33 0.44 0.38 0.36 0.49
Hurst parameter 0.89 0.89
Frame size (bits)
4
x 10 8
1. FIFO/ DropTail 2. Random Early Drop (RED) 3. Fair Queuing (FQ) 4. Stochastic Fair Queuing (SFQ) 5. Deficit Round Robin (DRR)
1
10 Mbps
7
6
2
44.736 Mbps
0.89 0.89 0.89 0.88 0.85 0.79 0.74 0.73
5
4
3
R
D
3
. . .
2
1
0 0 0.5 1 1.5 2 2.5 Frame number 3 3.5 x 10 4
4
n
Buffer size: 46 or 200 packets Packet size: 552 bytes
3. Packet loss:
Contribution of loss episodes (%)
10
2
10
1
• Simple loss statistics cannot capture complexity of loss patterns • We characterize packet loss using loss episodes • Real-time applications often more susceptible to consecutive packet losses • Loss episodes reflect the burstiness of packet loss Packets arrived at the router Loss episode of length 3
n n+1 n+2 n+3 n+4 n+5
FIFO/DropTail 46 packet buffer FIFO/DropTail 200 packet buffer RED 46 packet buffer RED 200 packet buffer
4. Wavelet analysis of packet loss:
• Traffic traces exhibit long range dependency (LRD) for time scales of 2 5 • Loss traces also exhibit LRD for time scales of 2 10 · 1ms ≈ 1.2s • The loss process capture the LRD characteristic of the traffic
·
40ms ≈ 1s.
10
0
10
-1
10
-2
10
-3
37
6 5 4 3 log 2 (Γ) 2 4 6 Octave j 8 10 12 2 1 0 -1
10
-4
0
2
4
6
8
10
12
14
16
18
36 35 34 33 log 2(Γ) 32 31
Loss episode of length 2
10
2
Length of loss episode (packets)
n+6
n+7
n+8
10 Contribution of loss episodes (%)
1
FQ SFQ DRR
30 29 28
Packets from flow 1
Packets from flow 2
10
0
27
2
4
6
8 Octave j
1 0
12
14
16
Successfully received packet Dropped packet Aggregate loss: Two loss episodes, one of length 3 the other of length 2 Per-flow loss: Flow 1: One loss episode of length 2 Flow 2: Two loss episodes, one of length 1 the other of length 2
10
-1
10
-2
Wavelet LRD estimator of 30-minute News traffic trace
0 5 10 15 20 25 30 35 40
10
-3
Wavelet LRD estimator of loss trace Buffer sizes: 46, 100, 200 packets Packet size: 552 bytes
10
-4
Length of loss episode (packets)
• The LRD behaviour is present regardless of the buffer size • These properties indicate self-similarity in loss processes
ASI Exchange March 12, 2002