IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
Scheduling TCP-Friendly Flows over a Satellite Network
R. Secchi, A. Sathiaseelan, and G. Fairhurst
Electronics Research GroupUniversity of Aberdeen
Aberdeen, United Kingdom{raffaello, arjuna, gorry}@erg.abdn.ac.uk
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
Outline
• QoS in Satellite Networking – System Requirements– Architecture Components
• Scenarios and Results– Criteria for performance evaluation– Relevant cross-layer interactions
• Conclusions & Future work
QoS in Satellite Networks
• Features & goals of the global Internet – Universal Interconnection– Decentralization– Heterogeneous Networking – Multiservice & IP oriented
• Objectives of QoS-Sat management– Integration in a large-scale network– Support of a wide range of applications
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
Approaches to QoS provisioning
• Limitations of hard QoS approaches (IntServ)– Traffic must be classified– A layer-3 to layer-2 mapping is required– Lack of network policies standardization: the end-to-end
behavior is difficult to guarantee– Subjectivity of policies: one provider’s gold traffic can be
bronze for another– The capacity argument
• End-to-end QoS schemes – High service levels under normal traffic conditions– Preventing collapse under high traffic loads
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
System Requirements
• A transport framework able to support multimedia applications
– Satisfying multimedia QoS requirements
– Compatibility with existing TCP services
– Network stability
DCCP
AQM, Fair queuing
BoD, Framing
• Active participation of IP layer– Fast feedback to upper-layer congestion control– Flow isolation & intelligent scheduling
• Effective use of satellite capacity – TCP/IP oriented design– Tuning knobs for efficiency vs. performance tradeoffs– Appropriate burst assignment across long
superframes
DCCP: The State of Art
• DCCP provides a framework for congestion control for multimedia flows– RFC 5348 (TFRC) proposed standard
• Suited for streaming media • Rate-based protocol• Requires sender and receiver participation• Periodic feedback reports• Smoothly variable sending rate
– RFC 4828 (TFRC-SP)• Small packet variant of TFRC• Suited for VoIP
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
TFRC development process
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
2000 20082006 2007
RFC 3448
RFC 4342
RFC 4828
RFC 5348
TFRC specifications
CCID3 in DCCP
TFRC-SP variant
Revision of TFTC
• Expanding TFRC framework (new CCID?)
• TFRC-SP in DCCP framework
future
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
SAL
MAC
IP
L4
Segmentation (ULE)
data signalling
Framing ATM, MPEG-2
flow control
To/From NCC
CR
buffer monitoring
DRADVB-Frames
TBTP
User applications
Transport (TFRC, TFRC-SP)
Packet SchedulingDrop Tail, DRR, SFQ, CBQ
Active Queue ManagementRED (Byte/Packet mode)
Bandwidth on DemandRBDC, CRA
VoIP packets
Architecture Components
Explicit/Implicit feedback
Bandwidth/Delay tradeoffs
Intra-flow & Inter-flow fairness
Simulation methodology
• Simulation scenario– Individual terminal scenario
– MF-TDMA over return link
– TDM over forward link
• Algorithms considered– DCCP: TFRC, TFRC-SP
– AQM: RED byte/packet mode
– IP layer: DropTail, DRR, SFQ, Prio
– BoD: RCDC, CRA
• Evaluation Criteria– R-score (objective quality estimation)
– Max-min fairness
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
Effects of flow isolation on TFRC
•A single TFRC streaming application may not achieve fairness when sharing with multiple data TCP flows• DRR is able to restore fairness and improve performance
Effect of AQM and Packet scheduling
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
Effects of Bandwidth on Demand
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
IWSSC 2008 – Oct 3rd 2008 Raffaello Secchi
Conclusions
• The soft QoS concept for satellite networks– Architecture definition and its purpose– Identifying main elements, their scope and cross-
layer interactions– Strong points and weaknesses
• Ideas for the future research– Parametric optimization of arch. elements– Towards an emulative testbed?