Reliable multicast is an essential network
service for supporting a class of dissemination-oriented
applications, such as file delivery, stock quotes, software
distribution, whiteboard sharing, distributed computing, which
is emerging in Internet recently. At the network level, IP
multicast provides an efficient one-to-many IP packets
delivery service but without any reliability guarantees and
congestion control mechanism. This project aims to design
efficient loss recovery and congestion control mechanisms to
support reliable multicast service in Internet.
In this project, we propose a new framework
which jointly performs Local Delivery and Congestion Control (LDCC).
In this framework, Delivery and Control Servers (DCSes)
collocated with routers perform LDCC functions. Each DCS and
its serving receivers form a local DCS region according to the
tree topology. With proper acknowledgement processing and
buffer management, packet loss can be efficiently recovered
locally, while the overall throughput degradation caused by
the interference of neighboring regions can be minimized by
“local” congestion control. We demonstrate through ns-2
simulations that our framework can achieve significantly lower
loss recovery latency without sacrificing the network
throughput, compared to existing approaches such as AER/NCA.
Next, we address the buffer management
issue in loss recovery schemes. We formulate the optimization
problem of caching policies as maximizing the amount of data
retrieved from the buffer. Based on the formulation, we
propose two algorithms of caching policy and show that they
can achieve near-optimal performance. Meanwhile, we analyze
the performance improvement of the two caching policies
compared to existing approaches. We also realize one of the
two near optimal caching policies and use ns-2 simulations to
demonstrate and validate its performance gain.
Since loss recovery and congestion control
are closely related issues, we further study the influence of
loss recovery schemes on the behaviors and performance of
congestion control mechanism. We identify the major
modifications to TCP reliability control (loss recovery)
mechanism in most of the RMCC schemes, including NAK-based
instead of ACK-based retransmission mechanism and local loss
recovery instead of end-to-end mechanism. We show that these
modifications may bring a number of substantial influences to
RMCC. We also show that with the modifications the steadystate
throughput of RMCC schemes can be accurately modelled without
considering the timeout effects. These findings and insights
could provide useful recommendations for the design, testing
and deployment of reliable multicast protocols and services.
In this project, we also investigate a key
issue in nominee-based congestion control, nominee selection
mechanism, which is essential for multicast services to ensure
fairness and congestion avoidance. Existing nominee selection
schemes choose nominees by comparing the calculated throughput
of receivers using the TCP throughput equation with the
measured loss rate and round-trip time. Since the calculated
throughput varies with different transmission rates, it may
not accurately indicate the eligibility of a receiver to be
the nominee. This causes the problem that a new nominee is not
necessarily “worse” than the current one and the “worse”
receiver could not be selected accurately. We address the
problem in existing schemes by identifying the conditions for
the valid use of calculated throughput. We propose a new
Generic Nominee Selection Algorithm (GNSA) as a solution and
prove that GNSA converges to the “worse” receiver and the
expected number of iterations is less than (1+In n),
where n is the group size. We demonstrate through ns-2
simulations the benefits of GNSA in terms of better fairness
properties and less iterations to converge than existing
nominee selection schemes such as that in TFMCC.
It is well recognized that QoS provisioning
in IP networks is a very challenging and hot topic. In this
area, we have made substantial progress. The objective is to
develop an efficient QoS provisioning mechanism for TCP/IP
network, which has good scalability and low implementation
cost. The new framework includes new scheduling algorithm,
admission control, congestion control schemes, buffer
management schemes and QoS routing protocol. So far, we have
developed a new packet scheduling algorithm, namely
Flow-state-dependent Dynamic Packet Scheduling (FDPS) that
realizes the Service Curves for guaranteed service
provisioning. In this scheduling scheme, admitted QoS-based
flows are protected yet allowing best effort traffic to
co-exist in a multi-services network. The proposed FDPS
algorithm performs packet monitoring, marking, scheduling and
discarding. With its fine granularity in packet marking, each
packet is forwarded and scheduled in a controlled and orderly
manner. We have shown through mathematical analysis that FDPS
algorithm defines the arrival and service curves required in
Service Curves and thus allows the theory of Service Curve to
be applied in a real network for per flow QoS provisioning. We
also gave an associated admission control example for this
guaranteed service. A tight end-to-end delay is also
guaranteed for QoS flows in a packet switching network.
Compared with existing scheduling algorithms for QoS
provisioning, FDPS can provide an integrated guaranteed and
best effort service in a multi-services network. We have
conducted extensive simulation experiments based on NS-2 and
the numerical results validate the theoretical analysis and
demonstrate the superiority of the proposed scheme. In
addition to the theoretical research work, we are also
developing a test-bed for the proposed QoS provisioning
mechanism; This testbed is based on programmable routers; The
new mechanism will have the following features: good
scalability; low complexity of implementation; low cost and
backward compatibility to current protocols.
Another research proposal is about active
packets for QoS control and traffic protection. It is proposed
to use active packets that dynamically adapt its QoS state to
varying conditions of the network and re-routing based on the
application QoS requirements for graceful QoS degradation
during the congestion state. The active packets require no
state maintenance in the routers and are topologically routed
to optimize on the available network resources. This possesses
the potentials for ensuring an optimal end-to-end QoS
guarantee without the complexity in the core routers. Through
proper QoS modelling and adaptation, it is expected that
end-to-end QoS can be ensured without over provisioning of
resources and traffic protection without significantly
sacrificing user’s QoS. This project identifies three focus
areas of research in QoS modelling, QoS adaptation and QoS
control.
