Advancement: Secure Efficient Real-time Data Transport in Hybrid Networks with Byzantine Failures

Speaker Name: 
Ramesh Srinivasan
Speaker Title: 
Ph.D. Student
Start Time: 
Friday, December 6, 2019 - 2:15pm
Location: 
Engineering 2, Room 280

Abstract

The original TCP design was intended solely for wired networks, which have since evolved into hybrid networks, comprising of co-existing wireless and wired technologies. There are three aspects that make the legacy TCP implementation inefficient in today’s hybrid networks. First, the implementation responds to packet losses and increased latencies due to wireless-link errors, as if it were due to network-congestion, resulting in reduced throughput and inefficiencies. Secondly, TCP responds to packet losses by retransmission of the packet after timeout expiration for the receipt of the ACK; this is a reactive response to the packet loss, resulting in end-end delays. Finally, particularly in 5G networks with mobile end-nodes, TCP proactively identifies compromised data (namely Byzantine faults), identifies the responsible intermediate nodes and links, and seamlessly recovers from the failures.

 

Accordingly, this thesis will focus on the design and analysis of several protocols for secure, efficient, and reliable real-time delivery of information in hybrid networks subject to Byzantine failures. We propose three protocols to achieve this. First, we propose the Wireless Santa Cruz Real Congestion Identifying Protocol (WSC-RCIP) to differentiate true network congestion from link-layer errors. The protocol keeps track of the number of link-layer retransmissions for each datagram, and introduces a mechanism for eliminating those times from the observed RTT for use as a measure of congestion in the network. Secondly, we propose the TCP Network-Coding Window Transformation protocol (TCP-NWT) to use network coding to proactively address packet loss without retransmissions along, with (TCP-NBD) NC with Byzantine fault detection (TCP-NBD), which integrates network coding with consensus mechanisms to detect and correct errors, as well as detect those nodes responsible for inj! ecting errors and TCP Network Nodes Confidence Rating Protocol (TCP-NCR) for evaluating confidence rating factors based on those nodes suspected of injecting errors.

Event Type: 
Adancement/Defense
Advisor: 
Professor J.J. Garcia-Luna-Aceves
Graduate Program: 
Computer Engineering Ph.D.