Michele Pagano – HET-NET Ilkely, July Modeling the Behavior of a DVB- RCS Satellite Network: an Empirical Validation Davide Adami, Stefano Giordano, Michele Pagano, Raffaello Secchi Dipartimento di Ingegneria dellInformazione Universita di Pisa
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July 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 bandwidth allocation end-to-end connection
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July RBDC Allocation Strategy 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 RBDC max and at least the Committed Information Rate (CIR)
Michele Pagano – HET-NET Ilkely, July Continuous Time Approximation (1) Let define … Instantaneous input rate measured at traffic station Rate assigned from RNCC to TT If less input rate is less than RBDC max 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 Requested bandwidth from TT to RNCC The requested bandwidth is a smoothed version of input rate. Assuming high time constant r(t) = x(t)
Michele Pagano – HET-NET Ilkely, July Continuous Time Approximation (2) By applying heavy load approximation to Lyndleys recursion, we have Thus, assuming packet buffering only at traffic terminal The queue size evolution is obtained by substituting previous expression and integrating
Michele Pagano – HET-NET Ilkely, July Satellite Measurement Test-Bed Total Capacity (Ct) 2122 Kbps Frame Period ms slot bandwidth (Cs) 44 kbps Resource Allocation Period 400 ms TT SAT SKYPLEX data terminal Ethernet LAN A B DBV-RCSDBV-S Characteristics of satellite link Laptop PC Since 18 TT were active, the available bandwidth was
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July UDP Traffic Measurements Packet Size (bytes) directionmean RTT (ms)std. dev. (ms) 1024 A-B-A B-A-B A-B-A B-A-B The path is symmetrical and introduces low jitter (2%) The behavior weakly dependent from packet size We observe very low drop rate (<10 -4 )
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July Average TCP Congestion Window W max = 32KB Slow-start exponential increasing advertised-window saturated three-way handshake TCP is unable to saturate channel capacity !
Michele Pagano – HET-NET Ilkely, July 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
Michele Pagano – HET-NET Ilkely, July 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