Page 1
1 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Modeling the Behavior of a DVB- RCS Satellite Network: an Empirical
Validation
Davide Adami, Stefano Giordano, Michele Pagano, Raffaello Secchi
Dipartimento di Ingegneria dell’InformazioneUniversita’ di Pisa
Page 2
2 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Outline
• Motivation
• Introduction to DVB-RCS architecture
– satellite network elements
– bandwidth allocation strategies
• The modeling methodology
• Modeling Validation through actual traffic
measurements
– impact on UDP Constant Rate Traffic
– behavior of TCP Short Lived Flows
Page 3
3 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Motivations
• Satellite Networks provide access to vast regions at a low cost:– The satellite link bandwidth is a scarce resource and its
use should optimized – The DVB-RCS is designed to support user interactivity
from satellite link and integrate satellite networks into the global Internet infrastructure
• An analytical framework is required:– To evaluate the behavior of BoD algorithms at IP layer– To evaluate the impact of satellite MAC on TCP/IP traffic– To the project of new satellite access scheme
Page 4
4 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
DVB-RCS Architecture
• The Regional Network Control Center (RNCC) provides control and management to a group of terminals
• The MAC allocation is based on Multi-Frequency Time Division Multiple Access (MF-TDMA) scheme
bandwidth
request
bandwidthallocation
end-to-end connection
Page 5
5 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
DVB-RCS Allocation Strategies
• Constant Rate Assignment (CRA)– Bandwidth is negotiated between the traffic terminal and
RNCC at the beginning of connection
• Rate Based Dynamic Capacity (RBDC)– Traffic terminals submit to RNCC bandwidth request
messages based on rate measurement of local incoming traffic
• Volume Based Dynamic Capacity (VBDC)– Each terminals request the amount of bandwidth per frame
needed to empty its buffer
• Free Capacity Assignment (FCA)– No explicit requests comes from terminals. The RNCC assign
bandwidth using some fairness criteria
Page 6
6 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
RBDC Allocation Strategy
11 krkxkr
The requested bandwidth is the smoothing of amount of traffic seen during k-th resource allocation
BoD controller
Traffic Terminal
BoD Processing
system response time (L frames)
safe frame period
k-th resource allocation period
r(k) a(k)
The BoD Controller assign bandwidth as long as is lower than the ceiling threshold RBDCmax and at least the Committed Information Rate (CIR) max,,maxmin RBDCCIRLkrka
1 2 3 4 5 6 7 8 9 10 11 12 13
14 15 1617 19 20 21 22
Page 7
7 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Continuous Time Approximation (1)
tx
ta
tLtxta Fs
Let define … Instantaneous input rate measured at traffic station
Rate assigned from RNCC to TT
If less input rate is less than RBDCmax and enough bandwidth is available, the bandwidth reserved for a traffic station is given by
The F is a positive noisy term that takes the discrete nature of time-slot allocation into account
trRequested bandwidth from TT to RNCC
The requested bandwidth is a smoothed version of input rate. Assuming high time constant r(t) = x(t)
Page 8
8 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Continuous Time Approximation (2)
tatxtatxtq tq 01
By applying heavy load approximation to Lyndley’s recursion, we have
Rta
RtqTtRTT
0
Thus, assuming packet buffering only at traffic terminal
t t
Fs ddLxxtq
The queue size evolution is obtained by substituting previous expression and integrating
Page 9
9 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Satellite Measurement Test-Bed
Total Capacity (Ct) 2122 Kbps
Frame Period 273.3 ms
slot bandwidth (Cs) 44 kbps
Resource Allocation Period 400 ms
TT TT
SAT
SKYPLEX data
terminal
Ethernet LAN
A B
DBV-RCS DBV-S
Characteristics of satellite link
Laptop PC
Laptop PC
Since 18 TT were active, the available bandwidth was
KbpsCCRBDC st 136417max
Page 10
10 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Measurement Sessions
• UDP Traffic Measurements– We use constant rate UDP traffic to evaluate the
characteristics of satellite link and validate our RTT model
• TCP Traffic Measurements – We schedule a new TCP connection carrying 600KB every
60 seconds and we evaluate the mean behavior of TCP cwnd and RTT
– Since the TCP connection does not meet losses, TCP never from slow-start phase and the ssthresh remains unset
– A simple rate profile is a rate pulse with exponential increasing
Dt
Rt
R
w
R
0max
1
1,2
min
Page 11
11 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
UDP Traffic Measurements
Packet Size (bytes)
direction mean RTT (ms) std. dev. (ms)
1024A-B-A 607.71 13.28
B-A-B 596.50 17.69
512A-B-A 601.35 12.40
B-A-B 590.07 15.19
• The path is symmetrical and introduces low jitter (2%)
• The behavior weakly dependent from packet size
• We observe very low drop rate (<10-4)
Page 12
12 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
UDP Flow Measurements: Throughput
The output presents an evident overshot after the transition time.
This phenomenon should be attributed to presence of non-linear term F.
Comparison between experimental and theoretical throughput
Page 13
13 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
UDP Flow Measurements: RTT
After the transition time the RTT undergoes a drastical increase. The extent of RTT increasing is inversely proportional to rate
Comparison between experimental and theoretical RTT
Page 14
14 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Average TCP Congestion Window
Wmax = 32KB
Slow-start exponential increasing advertised-window
saturated
three-way
handshake
TCP is unable to saturate channel capacity !
Page 15
15 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
TCP Round Trip Time Dynamics
• The theoretical RTT shows a good match with the real RTT dynamic
• TCP packets may experiment nearly two times the RTT observed during steady state period
We fed the RTT model with actual traffic data and compare the results with experimental RTT
Page 16
16 Michele Pagano – HET-NET 2005 - Ilkely, July - 2005
Conclusions
In this work, we evaluate the impact of BoD mechanisms on TCP/IP traffic by means of an analytical approach
Our analysis highlights some issues:– large delay variations determine long delays and
performance degradations– Short-lived TCP connections may achieve low
throughput due to RTT increasing during connections start-up
– The advertised-window allow 64KB at most, but the satellite link bandwidth-delay product is higher than that