Download feamster:education2011.pdf

Contents : Getting Students Hands Dirty with Clean-Slate Networking Nick Feamster Jennifer Rexford Georgia Tech Princeton University ABSTRACT Conventional networking courses treat today s protocols and mechanisms as xed artifacts rather than as part of a continually evolving system. To prepare students to think critically about Internet architecture we created a graduate networking course that combines clean slate networking research with hands-on experience in analyzing building and extending real networks. Our goal was to prepare students to create and explore new architectural ideas while teaching them the platforms and tools needed to evaluate their designs in practice. The course with offerings at both Georgia Tech and Princeton focused on network management as a concrete way to explore different ways to split functionality across the end hosts network elements and management systems. The programming assignments exposed students to a range of systems including Click Quagga Emulab OpenFlow/NOX Mininet the Transit Portal and publiclyavailable Net ow and BGP measurement data. 1. Introduction Most networking courses are organized around the layers of the network protocol stack implicitly assuming that the Internet architecture is set in stone. Network architect John Day argues that we teach the design of the Internet as a paragon of engineering rather than a snapshot of our understanding at the time . Historically this mode of instruction was a natural reaction to the dif culty in changing the form and function of existing protocols and networking equipment. Recent trends in networking research notably the clean slate networking philosophy and the increasing availability of programmable network elements create new opportunities to re-think the Internet architecture. Changes to the Internet could range from minor tweaks to a major refactoring of network functionality. These trends suggest to us that networking should be taught in a fundamentally different way but they also raise questions about how to structure a course without the familiar scaffolding of the network protocol stack. Because so many networking concepts are abstract so we want to present these concepts in a real practical context and equip students to go beyond high-level architectural ideas to speci c designs and implementations. In other words the course should both train students in clean-slate design and get their hands dirty with the details of network protocols and systems. Our new course design aims to achieve these goals through a combination of lectures selected readings and programming assignments that focus on speci c networking problems rather than any particular protocol layer or application. Organizing around problems rather than layers avoids redundant material while stressing concepts that appear at multiple layers of the protocol stack. For example the Ethernet back-off mechanism and TCP congestion control are both examples of distributed resource-allocation algorithms running on the end hosts. This organization also allows us to compare and contrast multiple approaches to the same problem. For example should load balancing over network paths happen through central optimization of the routing-protocol con guration or distributed control where hosts shift traf c to less-congested paths To scope the material our rst offering of the course focused on problems that arise in network management covering both how to manage today s networks and how to design future networks that are easier to manage. We explore a range of settings including home enterprise data-center and backbone networks. A cornerstone of the course is the set of hands-on programming assignments and laboratories that teach students the platforms tools and data sets commonly used in networking research. Our goal is to train students to take their architectural ideas beyond paper designs by building and deploying real systems. We also believe that handson experience is a better way to learn the considerable domain details in today s networking protocols than marching through the individual protocols in lecture. In the rst offering of the course the assignments covered programmable network elements (e.g. Click 7 Quagga 11 and OpenFlow/NOX 3 ) experimental platforms (e.g. Emulab 2 Mininet 8 and the Transit Portal 13 ) and measurement data (e.g. NetFlow traf c traces and BGP routing data). Using multiple platforms each with its own scripting languages and parsing tools also gave the students the con dence to learn new systems for their own research. We have taught two instantiations of graduate-level networking courses following the philosophy outlined above at both Princeton and Georgia Tech and course reviews have shown that students nd this new course design provides them both relevant practical experience and teaches them critical thinking skills th

INFO HASH : 0f1b185f21569e89579195dc8a9735822c0c0747

Download now

File Size: 101.7 kb