| |||||||||||||||||||||||||||||
Overview | Results | Publications | Software | People | References Overview:Several factors affect the overall performance of TCP in a gigabit wireless LAN. In this project we seek to investigate these factors with a view to proposing a better transport protocol for such a gigabit wireless LAN. The project is a part of the overall Yamacraw initiative to develop a working prototype of a 1Gbps wireless LAN. The project will have three dimensions. We will examine TCP performance in gigabit networks in general. We will also investigate the behavior of TCP on a MAC layer with bandwidth guarantees. Lastly, we will look at the TCP performance in a rapidly varying physical channel, characteristic of the Wireless LAN environment. Simulations will be performed and observations characterized with a view to proposing a better transport protocol for the environment under consideration. We will also implement our results in a test-bed to validate simulation results.
Results / Status:ResultsIn the first phase of the project, we investigate the performance of TCP over networks that provide Quality of Service (QoS) guarantees with a view to modifying the congestion control mechanisms of TCP to take advantage of bandwidth guarantees provided by the network. We have simulated the behavior of TCP in networks with guaranteed bandwidth and observed that since TCP is unaware of the underlying bandwidth reservations provided by the network, it fails to efficiently utilize the reserved bandwidth provided by the network. Figure 1 shows the topology used for our simulation. There are 6 flows, three of which have 4, 2 and 1 Mbps reservations respectively while the remaining three are best-effort flows.
We propose GTCP, which includes modifications to default TCP's startup and congestion control mechanisms to leverage bandwidth reservations provided by the network. Specifically, TCP is made aware of the reserved bandwidth at startup. Also TCP's fast recovery algorithm is modified to improve performance especially in networks with high loss rates. The modifications implemented in GTCP significantly improve the achieved throughput, compared to default TCP's performance. This is as shown below in Figure 2a below:
Figure 2b shows that GTCP achieves the expected instantaneous compliant throughput for each of the reserved flows.
We also investigated how GTCP scales with increasing network capacity, round-trip time, number of flows with reservation and the amount of reservation and our results show that GTCP indeed scales well.
StatusAt the moment, we are working on implementing the TCP with the new changes (GTCP) on the Yamacraw Wireless LAN prototype (as shown in Figure 3a), which has a QoS-enabled, TDMA-TDD based Medium Access Control (MAC). This essentially involves interfacing (Figure 3b) and network programming and aims to demonstrate the gains of GTCP over default-TCP in a QoS enabled MAC.
Furthermore, we are investigating the effects of lost acknowledgements on
the reverse path, different data paths for marked/unmarked packets and losses for
marked packets (controlled load service).
Publications & Presentations:
Software Downloads:For our simulations we are using ns-2 simulator version 1b8a, which is available on the web. Requests for actual scripts of our simulation should be sent to gte292x@prism.gatech.edu.People:
References & Related Work:TCP in Gigabit Networks
TCP over Wireless Networks / LANs
TCP over Networks with Guaranteed Bandwidth
|