Predictable
Delivery for Large-Scale Real-Time Applications
Predictable
real-time delivery is called for by many large-scale real-time applications
that cannot be effectively supported by existing solutions, e.g., they require large-scale
predictable (or soft) real-time delivery without reservations, and are deployed
across multiple administrative domains on dynamic networks.
In this project, we
will investigate the fundamental issues in supporting large-scale predictable
real-time delivery on the Internet. We will emphasize predictability at the
network layer. By predictability, we mean that, with a high probability, a
packet will reach its destination within a given delay. We will explore new
architectures and promising solutions. The fundamental question that we like to
answer is: Can large-scale predictable real-time delivery be achieved in a
scalable and efficient fashion? We will also emphasize dynamic network
conditions (e.g., multi-path) and incremental deployment.
Different from traditional QoS solutions, which are often limited to a single domain,
we consider a globally synchronized time available on routers to enable their
distributed collaboration across multiple domains. We believe that this
potential capability of the future Internet can help us develop effective
solutions to meet the challenge and greatly improve the current solutions. The
basic idea of the proposed framework is summarized as follows. Assume that a
packet of a real-time flow carries its birth-time t0, its delay requirement D0,
and its predictability requirement P0, i.e., it is expected to arrive at its
destination with a delay no longer than D0 with a probability P0 or higher.
With the synchronized time and the packet’s birth time, a router easily obtains
the upstream elapsed delay that the packet has spent on its path up to this
router. Then, based on the estimated downstream path delay and load information
from this router to the packet’s destination, the router can determine the
service priority for the packet. When a packet experiences more delay in its
upstream, its downstream routers automatically increases its service priority
to compensate the extra delay. When each router on the path collaborates in
this way, we are able to develop effective schemes to ensure a packet to reach
its destination on time within the given probability.
In this project, we will investigate
how the synchronized time can help us develop efficient and scalable solutions
to support predictable real-time delivery. To the best of our knowledge,
although synchronized time has been used in several settings such as sensor
networks, none of them has thoroughly examined such an idea in a large-scale
dynamic setting. The synchronized time relieves us from static reservations and
presents us more chances to address scalability, network dynamics, and
incremental deployability. To realize these
opportunities, we will develop novel mechanisms for obtaining sufficiently
accurate downstream delay estimation, and enabling routers to exploit the delay
estimation to achieve efficient predictable delivery and compensate estimation
error terms.
Principal
Investigator: Yingfei Dong
Research Assistant: Chi Fong Kuan, Ken Oyadomari
Alumni: Yu Jade Cheng,
Danny Lee
This project is supported by NSF CNS-1018971, NSF CNS-1127875.
Note: Any
opinions, findings, and conclusions expressed on this web site are those of the
author(s) and do not necessarily reflect the views of the National Science
Foundation (NSF).