r57 - 20 Nov 2008 - 03:08:26 - Main.sjanakiYou are here: TWiki >

Welcome to Networking Research Lab (NRL)

Networking Research Lab (NRL) is led by Injong Rhee along with a number of graduate students and research scientists. We study network protocols that enhance the well-being of computer networks, particularly, what we call today the Internet. In the evolution of the Internet, we have observed the birth and demise of many different types of computer networks, including ATM, Ethernet, optical networks, wireless networks, sensor networks and delay (or disruption) tolerant networks. With the evolution of new network technologies, network protocols must also adapt to and evolve with the new technologies. The mission of NRL is to enable this evolution of network protocols. We build protocols that are fast, scalable, adaptable, realizable and deployable. In the midst of these efforts, we discover fundamental properties of computer networks involving humans as the major enablers, which makes the research work all the more interesting. Our approaches are fundamental, but at the same time highly practical; we pursue lasting work that can help network users of today as well as of future.

Networking Research Lab (NRL) is led by Injong Rhee along with a number of graduate students and research scientists. We study network protocols that enhance the well-being of computer networks, particularly, what we call today the Internet. In the evolution of the Internet, we have observed the birth and demise of many different types of computer networks, including ATM, Ethernet, optical networks, wireless networks, sensor networks and delay (or disruption) tolerant networks. With the evolution of new network technologies, network protocols must also adapt to and evolve with the new technologies. The mission of NRL is to enable this evolution of network protocols. We build protocols that are fast, scalable, adaptable, realizable and deployable. In the midst of these efforts, we discover fundamental properties of computer networks involving humans as the major enablers, which makes the research work all the more interesting. Our approaches are fundamental, but at the same time highly practical; we pursue lasting work that can help network users of today as well as of future.

Research Highlights

* Congestion Control for Multi-hop Wireless Networks (putting differential backpressure into practice).

Checkout out our Youtube Videos for more information :

Starvation in Wireless Multihop Networks An Analysis
This experiment tries to find out the dominant cause for starvation in the flow in the middle problem
using TFRC.

A TFRC flow is started from nodes sn2e19 to sn2e10. Two more flows are introduced later. As the
traffic starts from these flows, sn2e19 - sn2e10 degrades to a throughput of 0.

The cause for the starvation of this flow is queue overflow as the intermediate nodes that are forwarding
packets of sn2e19 - sn2e10 are now unable to flush packets fast enough. We obtain the proportion of
packet drops due to hidden terminals and queue overflows. It can be seen that queue overflow is the
dominant cause for packet loss and hence starvation.

Unfair Contention in the medium results in a queue buildup and this leads to the eventual starvation of
the flow as the congested node is never able to flush packets fast enough.

The video shows the testbed with 3 flows. The long red line in the middle is the flow sn2e19 - sn2e10.
The graph on bottom left shows the instantaneous throughput of the flows. It can be seen that as the
new flows are introduced, the throughput of sn2e19 - sn2e10 drops to zero.

The graph on the right shows the statistics about hidden terminal errors (green bars) and queue overflows
(red bars). It can be seen that queue overflow error spikes indicating unfair medium sharing at the intermediate nodes.

Starvation in Wireless Multihop Networks A Soluion
We repeat the experiment with a flow in the middle scenario but this time using DiffQ TCP. DiffQ TCP is the
new congestion control algorithm that we have developed.

DiffQ is a generic congestion control framework which is based on differential backlog based backpressure.
Differential Backlog based backpressure was proposed in the context of utility maximization by Tassiulas
and Ephremides. DiffQ TCP is the direct application of these concepts in practice.

DiffQ TCP relies on source rate control, flow scheduling and MAC prioritisation. Source rate control is done
using backpressure. The flow scheduling at each node selects the flow with the maximum queue differential
and schedules it for transmission. MAC priority is assigned based on the differential backlog.

In this scenario, the flow sn2e19 - sn2e10 gets lesser access to the medium as before. This results in a
queue build up at intermediate nodes. While TFRC suffered starvation because of queue build up, DiffQ TCP
flushes the queues quickly due to flow scheduling and MAC prioritisation. MAC prioitisation ensures that the
flow which is suffering unfairness gets preferential access to the medium.

We do not observe queue overflows with DiffQ TCP which points to the effectiveness of the differential
backlog scheme in practice.

TCP Testing

BIC/CUBIC

Human Mobility

News

Date	Description
2008/12/17	Photos of Ajit's diploma ceremony are posted. Click here
2008/11/15	Lab photos are updated. Click here to see the pictures.
2008/10/14	Congratulation! Our MS graduates, Puneet Arora and Sanidhya Khilnani are going to join Microsoft.
2008/04/28	The Cisco Collaborative Research Initiative donated three Cisco 7600 routers for our researches. See more details
2008/04/11	Seong ik Hong to present his research work on Mobility models
2008/04/04	Jeong Ki Min to present his work on Wireless Mesh Networks for the Written Preliminary Exam
2008/04/01	Sangtae Ha, Ajit Warrier, and Injong Rhee attend the Stanford workshop on The Future of TCP: Train-Wreck or Evolution ?
2008/03/28	Lab Seminar Presenter: Prashant Baronia Paper Title: On Exploiting Asymmetric Wireless Links via One-Way Estimation, MobiHoc 2007 Room: 3211 EB II Time: 10AM
2008/03/14	Lab Seminar Presenter: Chisung An Paper Title: Structured streams: a new transport abstraction, ACM SIGCOMM 2007 Room: 3211 EB II Time: 10AM
2008/03/04	Sangtae Ha and Injong Rhee attend the ICCRG meeting and PFLDnet 2008 held in Manchester, UK.

More ...

Resources

Documentation: TWiki Reference Manual
If you want to test some page or twiki syntax, please use SandBox

This site has been accessed : 337 times.

Topic attachments
I	Attachment	Action	Size	Date	Who
jpg	Ajit_farewell2.jpg	manage	465.9 K	13 Nov 2008 - 01:51	Main.jkmin
png	Ncsu-testbed2.png	manage	21.4 K	01 Aug 2008 - 19:57	Main.sjanaki
png	back-pressure.png	manage	11.3 K	01 Aug 2008 - 19:57	Main.sjanaki
html	csc573_schedule.html	manage	101.6 K	25 Aug 2008 - 16:30	Main.sjanaki
PNG	cubic_curve.PNG	manage	5.2 K	01 Aug 2008 - 19:58	Main.sjanaki
jpg	mobility.jpg	manage	16.5 K	01 Aug 2008 - 19:58	Main.sjanaki

Main

TWiki Tip of the Day

Disabling individual WikiWords
Prevent a WikiWord from being linked by prefixing it with an exclamation point. Example: To escape SunOs ... Read on

Webs
Main
Sandbox
TWiki

Copyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback