Modelling and performanceanalysis of transport protocols and queue policies in high-speed networks
Bufferbloat–the problem of persistently fulllarge buffers–is leading to an increase in queueing delays across the Internet.This problem necessitates the study of TCP in the end systems, and Active QueueManagement (AQM) at the routers. In this talk, we look at TCP-AQM as anend-to-end feedback system: a viewpoint that enables us to apply tools fromcontrol theory and nonlinear dynamics to these networks. In particular, weconsider models for Compound TCP (default in Windows OS) coupled with variousAQM strategies, such as Random Early Detection (RED), Random ExponentialMarking (REM), Proportional-integral (PI) queue policy, and apply analyticaltools to gain design insight into these systems. We extend these insights tonetworks with heterogeneous traffic including other TCP variants,non-responsive UDP flows, short-lived HTTP flows through packet-levelsimulations. Further, we discuss how such insight could lead to novel andsimple queue management policies for Internet routers. To complete the picture,we also consider some other forms of congestion control, such as (i) purelydelay-based TCP (FAST TCP), and (ii) bandwidth-and-delay based congestioncontrol (Google’s new BBR) and develop models for them. We will also discusssome open questions, and ponder upon how such analysis could help us deploy AQMstrategies in Internet routers incrementally.
Sreelakshmi Manjunath, IIT Mandi
Sreelakshmi Manjunath is with the faculty of School of Computing andElectrical Engineering, Indian Institute of Technology Mandi. She received herBE (Electrical & Electronics Engineering) from Visvesvaraya TechnologicalUniversity in 2012, and her MS and PhD degrees from Department of ElectricalEngineering, Indian Institute of Technology Madras in 2018. Before joining IITMandi, she was a postdoctoral researcher in the Department of Electrical andSystems Engineering at Washington University in St. Louis (MO, USA). Herresearch interests are communication networks, vehicular networks, controltheory and nonlinear dynamics.