Transcript CSE 5344 Computer Networks

An Integrated, Distributed Traffic
Control Strategy for Future Internet
H. Che
UTA
W. Su & C. Lagoa X. ke, C. Liu, & Y. Cui
Penn State
Tsinghua
Outline



Problems
Strategy
Conclusions
Problems
Limitations of the existing distributed traffic control solutions:

Solutions at different layers are developed independent of one another. As a
result, they may adversely interact with one another, attempting to achieve
conflicting design objectives [1][2]

They are largely empirical by design, without provable properties, such as
stability and optimality

The existing theoretical results have limited scope (e.g., single-domain, single
CoS, allowing limited number of design objectives). They cannot be used to
guide the protocol development to enable rich service quality features,
including Quality-of-Service (QoS), Traffic-Engineering (TE), and Fast-FailureRecovery (FFR)
Apparently, “patching” the Internet with add-on traffic control Features at different
layers independently is problematic
The aim of this work: to develop a strategy for integrated, multilayer
protocol development to enable rich service quality features at global scale,
including QoS, TE, and FFR
[1] L. Qiu, Y. R. Yang, Y. Zhang, and S. Shenker, ``On Selfish Routing in InternetLike Environments,“ ACM SIGCOMM'2003, Aug. 2003.
[2] Y. Liu, H. Zhang, W. Gong, D. Towsley, ``On the Interaction Between Overlay
Routing and Underlying Routing," IEEE INFOCOM'05
An Integrated Strategy
Outline:

A theoretical foundation

An integrated control structure
Theoretical Foundation
Idea: to make use of a distributed, QoS-aware, multipath forwarding paradigm
This forwarding paradigm is enabled by two large families of optimal, distributed controllers
(allowing unlimited number of design objectives, multipath, and multi-CoS):

end-to-end: require single-bit binary feedback, allowing pure end-to-end control at
transport layer

edge-to-edge: allow multi-domain edge-to-edge “per-hop” control at IP layer

An Internet access point performs single-hop control to enable CoS features for
CoS-based flow aggregates

A domain edge nodes performs CoS-agnostic control to enable TE and FFR
features for destination-based flow aggregates: inter-domain per-hop control
and intra-domain edge-to-edge control (with or without involvement of core
nodes for feedback control)
QoS-aware
end-to-end
control
CoS-agnotic
intra-domain
control
CoS-agnotic
inter-domain
control
CoS-aware access
control
Theoretical Foundation
Why the two families of controllers help:


They make it possible to develop distributed traffic control
protocols based on THEORY to enable rich QoS, TE, and FFR
features at global scale
They are highly scalable and can deal with tussles and network
diversities
Integrated Control Structure
Outline:


IP layer and overlay integration
IP layer and transport layer integration
Integrated Control Structure
IP layer and overlay integration
Goal: to minimize adverse interactions between overlay traffic
control and IP layer traffic control
Our Solution: let a network-based overlay service network involves
all the IP domain edge nodes under its coverage so that our
multi-domain control mechanism can be simultaneously applied
to both the IP layer and overlay in an integrated fashion
Integrated Control Structure
IP layer and transport layer integration:
Goals:

To minimize adverse interactions between IP rate adaptation for TE
and transport layer adaptation

To minimize the effect of IP rate adaptation for TE on transport layer
rate guaranteed flows
Solution:

Implementing three CoSs at IP layer: BE, AF with a target rate, and
an upper bounded rate service

All the adaptive end-to-end flows (e.g., TCP) are mapped to the upper
bounded rate service without call admission control

All the rate guaranteed end-to-end flows are mapped to the AF CoS
with call admission control

All the non-adaptive BE end-to-end flows (e.g., BE UDP) are mapped
to the BE CoS
Conclusions
Developed a strategy for traffic control protocol development
at multiple layers, possessing the following expected features:




They are integrated, achieving non-conflicting design objectives
they provide rich service quality features, including QoS, TE, and FFR
They can deal with network diversities and tussles
They enjoy provable properties such as scalability, stability, and
optimality
Caveat: The above expected features are derived from a theoretical
Framework based on a fluid-flow model. It is a work-in-progress. How
closely the protocols developed based on this strategy will achieve the
above expected features is subject to future investigation