470 likes | 687 Views
Transport Layer Design Perspectives for Heterogeneous Networks. Vassilis Tsaoussidis , Democritus University of Greece IEEE WLN 2006, Keynote Tampa, Florida, 14 November 2006. Keynote. A research monologue A critique of other people’s work A critique is easy – others do, you object
E N D
Transport Layer Design Perspectives for Heterogeneous Networks Vassilis Tsaoussidis, Democritus University of Greece IEEE WLN 2006, Keynote Tampa, Florida, 14 November 2006 Transport Layer Design Perspectives for Heterogeneous Networks
Keynote • A research monologue • A critique of other people’s work • A critique is easy – others do, you object • A critique is tough and risky – you object, they criticize you • World is conservative • and will remain so even after the keynote Transport Layer Design Perspectives for Heterogeneous Networks
Research Perspectives • Problem-driven research • Seek answers and then seek problems • Modify answers and then modify problems again • GOTO 3 • Is research today autonomous or is it driven by something (applications, industry etc.) Transport Layer Design Perspectives for Heterogeneous Networks
ANSWER • Yes, it is autonomous • Because it is research-oriented research Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Research Basics • Schedule data transmission of multiple flows • Schedule retransmission • Maximize: • Application Efficiency • through • System efficiency • System fairness Transport Layer Design Perspectives for Heterogeneous Networks
Network is a box • The box is black • The box is gray • The box is green Research tendency Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Research Primitives Application Efficiency Diversity System efficiency Calls for System fairness Impacts definition Priorities Diff Services Schedule Transmission/ retransmission requires impacts Max Utilization (goodput) requires Responsive behavior heterogeneity differentiates Requires enhanced Measurements, Detection requires Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Research Primitives Application Efficiency System efficiency System fairness Schedule Transmission/ retransmission • What can we measure • How precisely • What is fair – what is • our perspective requires Max Utilization (goodput) requires Responsive behavior Measurements, Detection requires Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Research Primitives • What do we really know • Granularity of measurements and network dynamics • Precision of detection • Correctness of response • Fairness Transport Layer Design Perspectives for Heterogeneous Networks
What do we know Transport Layer Design Perspectives for Heterogeneous Networks
From knee to cliff: What do we know • How fast do we (want to) go there? • We study one network dimension (space) to optimize another (time). Yet, we may study two dimensions together (space and time) and optimize both. From knee to cliff. • Responsiveness depends on contention • Can we really detect the knee? Transport Layer Design Perspectives for Heterogeneous Networks
Responsiveness vs. Buffering • Buffers help efficiency, delay notification • no, it’s not just scheduling really, it’s more than that. • AIMD Properties Notification points Transport Layer Design Perspectives for Heterogeneous Networks
Properties and Secrets of Grey Box.Let’s consider one more dimension: time • The granularity (and precision) of measurements • vs. • the granularity of changes Even if we find out, we just know for now Transport Layer Design Perspectives for Heterogeneous Networks
Congestion vs. Contention • Points of congestion–>points of synchronization • Timeout alleviates congestion – and prepares for a new round…. • Congestion Window – how much do we need it? • Does it capture contention as well? Transport Layer Design Perspectives for Heterogeneous Networks
Correctness of Response • Similar RTT measurements • Identical RTO settings • Multiple Simultaneous Retransmissions Transport Layer Design Perspectives for Heterogeneous Networks
Contention increase – timeout decrease! RTT RTO = A + 4D Transport Layer Design Perspectives for Heterogeneous Networks
Contention Decrease reverses the dynamics Transport Layer Design Perspectives for Heterogeneous Networks
Research directions • Contention is the key • Temporal characteristics of spatial properties is the anti-key • Traditional methodologies to gather information about the network may lack precision and appropriate granularity • Buffers delay detection – not just service Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Research PrimitivesHeterogeneity and Application Diversity Application Efficiency Diversity System efficiency Calls for System fairness Impacts definition Priorities Diff Services Schedule Transmission/ retransmission requires impacts Max Utilization (goodput) requires Responsive behavior heterogeneity differentiates Requires enhanced Measurements, Detection requires Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Research Issues • Application Diversity and Network Heterogeneity • Response strategy and cost of strategy • Service strategies, marking, priorities • Bottlenecks of marking • Fairness Transport Layer Design Perspectives for Heterogeneous Networks
End-to-end Transport deals with • World Wide Grids • Global Sensors • Moon and Mars • Typical Multimedia Transmission • TypicalTypical Internet Applications Transport Layer Design Perspectives for Heterogeneous Networks
What a design can do… • Increase application performance • Typically through system performance – there are exceptions! • Increase number of happy applications • Typically through multiplexing/fairness • Extend mobile device lifetime • Energy performance • Decrease cost • Reduce overhead, transmission effort Transport Layer Design Perspectives for Heterogeneous Networks
The control button Transport Layer Design Perspectives for Heterogeneous Networks
The control button Transport Layer Design Perspectives for Heterogeneous Networks
The control button Transport Layer Design Perspectives for Heterogeneous Networks
Application Performance • With a little help from our friends • Application- network- and device-oriented transmission and service patterns • Remember: Responsiveness and smoothness have different tradeoff dynamics in wireless networks • A Cross-layer collaborative scheme • See real-time video streaming Transport Layer Design Perspectives for Heterogeneous Networks
Binding Application Network and Device Dynamics Example: Scalable Streaming Video Protocol (SSVP): • an end-to-end congestion control scheme optimized for unicastvideostreaming applications • enables AIMD-basedcongestion control on top of UDP by properly adjusting the inter-packet-gap, spacing outgoing packets evenly to produce a smoothed flow • introduces new congestion control parameters adaptable to the prevailing network conditions • based on RTT measurements, automatically tunes βto prevent the system from operating below point knee (where bandwidth is underutilized) Transport Layer Design Perspectives for Heterogeneous Networks
Transmission tactics • With mobile, battery-powered devices • Aggressive or Conservative? • What is the corresponding relation of error-pattern and transmission tactic? • What is the impact of transmission tactic on system fairness • What is the relation between the risk to be excluded from service and the energy efficiency of the system Transport Layer Design Perspectives for Heterogeneous Networks
Impact of strategy on application performance Transport Layer Design Perspectives for Heterogeneous Networks
Goodput dynamics over heterogeneous networks Transport Layer Design Perspectives for Heterogeneous Networks
Note • No appropriate strategy so far Transport Layer Design Perspectives for Heterogeneous Networks
New perspectives • When network is heterogeneous, we care about the total impact on flow characteristics. • Splitting may fail to implement one single strategy • Response strategies. Distinct error patterns and causes may require similar responses • Detection and response strategies need to be associated with statistical confidence Transport Layer Design Perspectives for Heterogeneous Networks
Correctness of response • Conservative TCP transmits 6.3GB with 174.2MB overhead and consumes 300mAh energy • Aggressive TCP transmits 8,4GB with 1,5GB overhead and consumes 304mAh energy • NewReno transmits 6,1GB with 543MB overhead and consumes 303mAh Transport Layer Design Perspectives for Heterogeneous Networks
Service Risk vs. Efficiency Transport Layer Design Perspectives for Heterogeneous Networks
More on transport-layer energy efficiency • Transport-layer energy efficiency does not exist • Protocol effort exists (EEE) • Protocol effort does not suffice to characterize efficiency(UAR) • Protocol effort may be translated into energy expenditure – but may not (EP) Transport Layer Design Perspectives for Heterogeneous Networks
Simple sample metrics we need Energy Potential: The design that leads to zero EEE and UAR when the device is sufficiently sophisticated Transport Layer Design Perspectives for Heterogeneous Networks
Energy management • Energy management a cross-layer issue • If one layer lacks sophistication, the design goes wasted • However, a good design may work on one device and not in another • One layer drives others to increase sophistication Transport Layer Design Perspectives for Heterogeneous Networks
LIBS: Less Impact Better Service • About LIBS • LIBS and Fairness • System and user priorities • LIBS and Priorities • LIBS and Marking Transport Layer Design Perspectives for Heterogeneous Networks
Internetworking with sensors Transport Layer Design Perspectives for Heterogeneous Networks
LIBS • Small packets get better service – as far as their impact on other packets is not significant • Marking and LIBS: packets with the same mark may be serviced on the basis of their impact • Marking alone is static and the marking scale cannot always adequately represent the service scale that optimizes application performance for all applications in the system • Priorities is both guys business: application’s and system’s Transport Layer Design Perspectives for Heterogeneous Networks
LIBS perspective of Fairness Fairness is redefined on the basis of application characteristics but in a system-oriented manner; not a service-oriented manner. A service-oriented manner cannot result in a zero-sum game; Service requests may be conflicting. Transport Layer Design Perspectives for Heterogeneous Networks
Less Impact Better Service Transport Layer Design Perspectives for Heterogeneous Networks
Happy Users Transport Layer Design Perspectives for Heterogeneous Networks
Research Directions • Error nature distinction is half the story • Response strategy does not correspond to wired/wireless error • Smoothness and Responsiveness have different dynamics in wireless environments • Responsive strategies do not presently exist • LIBS: Sensors require a new thinking philosophy • Energy potential is not always translated into energy efficiency Transport Layer Design Perspectives for Heterogeneous Networks
Transport ResearchMyths and Bottlenecks • TCP can do all? Not really • Wired or wireless? Don’t care • Network State Detection cannot be associated with buffers size only • Buffers help scheduling – but delay rescheduling (cost on granularity) • Architectures that allow Upper Layer Sophistication? By definition, more sophistication harms the dummies. Cross-layer (miss-)collaboration • Layering on demand Transport Layer Design Perspectives for Heterogeneous Networks
Further conclusions • Focus on flow properties, driven by channel, device, application properties • Service is not a network function alone nor is it an application function alone. The network may override priorities and service requests • Measurements are useful as soon as their granularity exceeds or follows the rhythm of changes. The box becomes green when the network becomes local • Responsive behavior may be targeted only when detection and recovery strategies suffice • Extending the internet does not mean extending TCP Transport Layer Design Perspectives for Heterogeneous Networks
Relevant Papershttp://comnet.ee.duth.gr/comnet/ • V. Tsaoussidis and I. Matta,"Open issues on TCP for Mobile Computing”, The Journal of Wireless Communications and Mobile Computing,WCMC John Wiley & Sons, Issue 1, Vol. 2, February 2002 • V. Tsaoussidis and C. Zhang, “The dynamics of Responsiveness and Smoothness in Heterogeneous Networks”, IEEE Journal on Selected Areas in Communications, (JSAC), Special issue on Mobile Computing and Networking, vol. 23, no. 6, June 2005 • C. Zhang and V. Tsaoussidis,”TCP Smoothness and Window Adjustment Strategy”, IEEE Transactions on Multimedia, 2006 • L.Mamatas, V.Tsaoussidis,"A new approach to Service Differentiation: Non-Congestive Queueing, CONWIN 2005, Budapest, Hungary • Psaras, V. Tsaoussidis, "Why TCP Timers (still) Don't Work Well“, ComputerNetworks (COMNET), 2006 • P. Papadimitriou and V. Tsaoussidis "End-to-end Congestion Management for Real-Time Streaming Video over the Internet“In Proc. of the 49th IEEE GLOBECOM 2006, San Fransisco, USA, November 2006 • L. Mamatas, T. Harks and V. Tsaoussidis, “Approaches to Congestion Control in Packet Networks”, Journal of Internet Engineering, Klidarithmos Press, January 2007 Transport Layer Design Perspectives for Heterogeneous Networks