CS 268 Semester Project

Project Teams and (Tentative) Project Titles

• Mosharaf Chowdhury, Sameer Agarwal, Packet Classification Using Explicit Coordination
• Matthias Vallentin, Evaluation of 802.11n for Long-Distance High-Bandwidth Multi-Hop Links
• Sara Alspaugh, Laura Keys, Andrew Krioukov, Datacenter Networking and Energy
• Priyanka Reddy, Bulk Transfer Protocol within Multiple Client-Server Networks
• Adrienne Porter Felt, Tier Splitting for Network-Bound Mobile Applications
• Neil Conway, Improving Hadoop Performance
• Nick Lanham, Charles Reiss, A Framework for Network-Aware Applications
• Arash Parsa, Minimal Interfering Wireless Networks
Projects from last semester can be found here.

The Essential Five Questions That Any Project Proposal Should Answer (The Heilmeier "Catechism")

• What is the problem you are trying to solve?
• What is the state-of-the-art and where does it fall short?
• What is your initial approach?
• What resources do you need for success?
• What is your rough timeline of work?

Project Schedule

• Week 21 September: Initial project proposal meetings with the instructor
• Th, 24 September: Short In-class Project Proposal Presentations
Single Quad Chart (Example: [ppt] [pdf])
• Northwest Quadrant: Project Graphic, Project Name, Project Team Members
• Northeast Quadrant: What is the research contribution of the project? Bullet points on the new and interesting concepts to be investigated in the project.
• Southwest Quadrant: Why do we care if you are successful? Bullet points on the potential impact of the project.
• Southeast Quadrant: Project milestones and schedule through the end of the semester.
• Week of 12 October: Second round project meetings with the instructor
• Th, 22 October: Checkpoint In-class Project Status Presentations
Revised Quad Chart
• Northwest Quadrant: Project Graphic, Project Name, Project Team Members
• Northeast Quadrant: Revised statement of the goals of the research project: what is its research contribution? Refined bullet points on what is to be investigated in the project.
• Southwest Quadrant: Bullet points on issues resolved and remaining open issues to be addressed.
• Southeast Quadrant: Updated project milestones and schedule through the end of the semester, highlighting accomplishments to date.
• Week of 9 November: Third round project meetings with the instructor
• Tu, 8 December: Project Poster Session, 4th Floor Soda Hall, 3-4:00 PM
• Provide a short project abstract, 5 - 10 sentences describing the problem addressed by your project and your results, by 7 December 2009. These will be made available on the web site and used in emails to advertise the poster session.
• M, 14 December: Final Project Write-ups due at 1700 hrs
• Projects reports are expected to be 5 pages long for one person projects, 10 pages for two person projects, and 15 pages for three person projects. These are only guidelines, not requirements or limits. You may submit a report that is as long as it needs to be to describe your project. Don't stretch the report just to meet a particular page number, nor feel constrained if it needs to be longer to get your key points and results across. You can add appendices to document details you wish to communicate to me. It is perfectly fine to submit your report as a wiki or set of web pages if you prefer.
• A good report must answer the standard questions about any research project. The following format is suggested:
  • Abstract: WHAT did you do, WHY was it important, WHAT were your high level results?
  • Problem Statement: WHAT is the problem you attempted to solve?
  • Prior Work: HOW have others approached your problem, and WHY were they only partially successfully?
  • Research Approach: WHAT was your approach to solving your problem? WHAT did you build? WHAT was your experimental methodology in terms of instrumentation, data collection, analysis, etc.? Include figures to describe your architecture and to assist in the presentation of your algorithms and analyses.
  • Results: WHAT were your results? HOW did you evaluate your work and WHAT were your figures of merit? Include graphs, charts, or other ways to visually describe your results.
  • Lessons Learned and Future Work: If you knew then what you know now, WHAT would you do differently? WHAT questions are left for future research?
  • Summary and Conclusions: Summarize your project and place your results in an overall context.
  • You may add appendices to document any additional material you wish to include with your report, such as code, API definitions, or system architecture block diagrams.
  Please note that the bulk of the report should be Sections 4 and 5.

Initial Project Ideas

• Datacenter and Enterprise Networking: Datacenter and non-University enterprise packet traces are not relatively available, and if they existed, they would benefit a large number of design studies. Collecting such traces is non-trivial. Good application level generators need to be selected, and you need access to a representative datacenter or enterprise. It may be possible to pursue such a project with one of our collaborating companies, such as HP or Yahoo!. You can access the Department's R-cluster or DeterLab for experiments, but the number of nodes is modest. We can arrange for access to Amazon EC2, but this is a virtualized environment, and traffic not visible to the trace collector can impact network behavior. One of the Open Cirrus Clusters might offer a better collection environment. Once packets are collected from a variety of places in the network, correlating activity sourced from multiple points is a problem of its own. However, being the people who collected such traces could make you very famous.
• Datacenter Networking: A number of recent proposals--two of whom we will study in this class--propose alternatives to the traditional tree-structured network topology. For example, the revered computer design Chuck Thacker has proposed an alternative datacenter interconnection scheme, based on hypercubes, but still using commodity ethernet technology for the internode links (1 G, 10 G, 40 G). These multicomputer interconnects, widely studied in the late 1980s, yield higher cross-section bandwidth at the possible cost of a larger number of hops between nodes. This project would involve analyzing these tradeoffs in the context of the communications patterns of typical datacenter applications. This is a good project for a team interested in both architecture and networking.
• Datacenter Networking and Power Reduction: Networking equipment remains the highest power consumed per square foot of machine room space in the datacenter, and remains the area of datacenter architecture that has benefited the least from power management schemes. Furthermore, the network is known (at least as a folk wisdom) to be underutilized and to exhibit bursty access. A project in this area would investigate mechanisms for reducing network power, by reducing the number of physical ports and links in the network, by selectively shutting them off. This could be driven by models of predictive models of network usage that indicate that the network will enter a period of relatively low activity. A key challenge here is the mechanism for determining when to shift between the on and off state and how to adjust the subset of the network to be turned off. An issue is how to reestablish not only routability as quickly as possible following powering up a previously sleeping portion of the network. Clever use of proxies, for "keep-alive" and other protocols, will need to be designed. This could be a good project for students with a mathematical bent teamed up with implementation-oriented students.
• Datacenter Networking for Computation Grid and Web: A question arises--at least to me!--as to whether the underlying network access patterns for datacenters dedicated for supercomputing applications or for large-scale web site data analytics are similar or different. Could you develop network topologies optimized for one over the other, or come up with a scheme that supports both reasonably well? Unlike the trace study suggested above, this project is more of a design study combined with an analysis of alternative network topologies and technologies. It will be important to characterize the network patterns that underlie the two different classes of applications. This could be a good project for a student who is interested in the integration of traditional grid and cloud computing.
• X-trace Enabled Routers: George Porter and Rodrigo Fonseca developed X-trace, a framework for using packet paths to collect information about network applications. It would be interesting to be in a position to use the framework below the sockets level inside network elements. This can be done, for example, by extending the CLICK software router to make it X-trace aware. Such an extended router could be useful, for example, to correlate packet streams from multiple sources for collecting whole network traces.
• Multicore Network Processing: Packet inspection and classification, for security, accounting, and quality of service, has often been rejected by the network research community as too expensive (in terms of latency). This project would examine how multicore processors could be used to handle pattern matching applications applied to high throughput network traffic. This is another project that might appeal to students interested in both architecture and networking (Prof. Keutzer in the ParLab is also interested in this topic).
• Multinetwork Interface Power: The following project was suggested to me by Bryan Lyles, a researcher at Bellcore. You have (i) multiple interfaces (wifi, wiMax, 3GPP, ....), (ii) each with different power saving modes, (iii) each with different capabilities in the modes, and (iv) each with different times to move between modes. The goal of the project is to represent power modes in a device independent interface, separate description/mechanism and policy, and enable policy control. Industry is trying to do this in 802.21. The documents can be found at https://mentor.ieee.org/802.21/documents You will probably want to look at: https://mentor.ieee.org/802.21/file/08/21-08-0243-00-mrpm-mrpm-sg-closing-report-july-2008.ppt and https://mentor.ieee.org/802.21/file/08/21-08-0230-01-mrpm-mrpm-redefined-scenarios-presentation.ppt since these contain pointers to a bunch of relevant presentations. We do not have a good methodology for representing the flexibility intrinsic in the various technologies.
• Campus Wireless Wardriving: It would be interesting (at least to me!) to make an extensive catalog of wireless LANs on and around the Berkeley campus, with an eye towards quantifying the interference issues, lost performance, and difficulty of cross-network mobility. For example, if you are authenticated for the EECS wireless LAN and AirBears, why can't we have tool that allows us to seamlessly move between these networks, automatically providing the necessary credentials on our behalf as we move from Soda Hall into the campus area?
• Virtual Machine Networking Support: This is a project that may be of interest to an OS student. There hasn't been much work that I am aware of that tries to manage the physical network interface that is shared among multiple virtual machines to ensure that it is used fairly. For example, can a single VM fill the buffers of the physical network interface, monopolizing its usage and starving other VMs' access to the network? This is a classical kind of research project in that first you would have to assess the fairness of virtual network access to see if it is a problem, propose a solution (perhaps scheduling of buffers), and then showing that your proposal results in improved network performance for the collection of VMs running on the same physical hardware.
• Network Security: We will be studying BotNets this semester as a special topic in network security, I would certainly consider projects in this area. Professors Paxson, Song or Wagner may be able to make some suggestions here.
• Wireless Networking: The car companies are very interested in enabling vehicle-to-vehicle communications, to enable, for example, cooperative collision avoidance systems. As an accident prone driver, I think this is a neat kind of network application that could led to an interesting project after checking some literature. How well do existing 802.11 wireless access point work at high velocities and rapidly converging angles? (Perhaps this is of interest to an EE communications person.) The communications must be reliable, even in the presence of challenging wireless conditions. There are some interesting privacy and security issues here, to make sure that the system can neither be subverted or tricked.
• Always a good idea to look at the last SIGCOMM conference proceeding or NSDI conference proceedings to get some ideas for next step opportunities in the work presented. For example, the last SIGCOMM has several data collection and analysis papers; not every paper proposed a new mechanism and evaluated it. This might suggest some ideas to you to pursue.

Writing Papers

Simulation

Testbeds and Emulation

The RON Testbed

Planetlab

Wireless Nodes

Routing

• The Click Modular Router
• XORP the eXtensible Open Router Platform
• Quagga - an open source routing suite

Measurement

• Scriptroute (note: runs well on planetlab and the RON testbed)

Analysis

• Books: Raj Jain, the Art of Computer Systems Performance Analysis - a very good overview of lots of mathematical techniques and queueing bits, aimed at a systems audience

Last updated: Tu Nov 26 15:16:55 PST 2009