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. The U.S. National Science Foundation

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).