Network Monitoring

• Bittorrent Lifetimes
  Bittorrent is a peer-to-peer file sharing tool that allows a file, seeded by a single source, to be downloaded concurrently by a set of users. It is interesting in that downloading users must also upload portions of the file to each other, lessening the load on the initial seed. In fact, the original source of a file may discontinue it's participation in the session once it has sent a single complete copy of the file. In this manner, a file may be made available over a long period of time without being completely available at a single location (as users who finish downloading files will typically quit the bittorrent session).

  This project is concerned with understanding how files shared in such a manner behave over time. It consists of two parts:

  Part A: You will read the bittorrent source and provide a description of how the protocol works. The intention of this model is to identify specific factors (i.e. the arrival/departure rates of users, access bandwidth of clients, file size, etc.) that effect how long an object will remain available. For instance, a large increase in access bandwidth might potentially break such a scheme as downloads would complete too quickly for a community of active clients to persist.

  Part B: You will validate your model by collecting real data on existing bittorrent sessions. This part will involve writing monitoring tools to watch bittorrent 'in the wild'. It will be important to attack this part of the project very early in the year, in order to have an opportunity to collect a reasonable amount of data. As this is likely to involve active monitoring (your monitor will connect to live torrents), careful design will be required address concerns such as the maximum bandwidth consumed.

  Thanks to Andrew Warfield for thinking of this one

Data Mining

• A Data Mining Approach
  

  Data Mining can involve the training of classification schemes to recognise desirable data and then operating these classifiers upon databases inorder to locate the data of interest.

  A useful data-mining framework already exists called Weka. However, Weka does not yet implement QDA (Quadratic Discriminant Analysis). This project would rectify the situation.

  Weka is writtern in java and has copiuos documentation describing its interfaces; what is required is that the student recognise the difficulties that the QDA algorithm posses and design an appropraite strategy to cope.

  An implementation of QDA (non-weka) is available for cross-validation.

  The link for Weka.

  Note: this project requires a student that is not scared by the java programming language and is comfortable with mathematics.

  QDA is described in Ripley, B. D. (1996) Pattern Recognition and Neural Networks. Cambridge University Press.

  Excellent materials covering both the (validation) implementation of QDA and the QDA technique itself are available in the University Stats Lab.

• Real-Time Data Mining
  

  Data Mining can involve the training of classification schemes to recognise desirable data and then operating these classifiers upon databases inorder to locate the data of interest.

  A useful data-mining framework already exists called Weka. However, Weka is not designed for Real-Time operation, this project would attempt to rectify this situation.

  Weka is writtern in java and has copiuos documentation describing its interfaces; what is required is that the student recognise the difficulties with real-time code implementations, particularly in Java, and design an appropraite strategy to permit use of this system.

  The objective would be to provide a library of operations that allow a subset of Weka operations to be incorporated into other code. Students will also need to validate their approach as to be real-time the approach must possess some time-bounded properties; the program cannot make a library call that might never return.

  The link for Weka.

  Note: this project requires a student that is not scared by the java programming language. A familiarity with C will assist in creating an appropriate library.

Operating System projects

The systems research group is continuing to develop and do research using Xen, but there are a couple of areas which we feel would make good self-contained student projects:

• DummyNet plus Xen = planetlab in a box
  The virtualization of operating systems provided by Xen allows us to intercept the interactions between an OS and the hardware devices that it uses. An example of this is to intercept packets to or from the network devices and overlay properties onto the packets as they enter/exit the network. Additionally, Xen can allow network traffic of one virtual machine to be exchanged with another virtual machine, through a virutal network interface. The properties of this virtual interface (such as the delay between machines) can also be changed. Such an ability --- to emulate change in the physical properties of the underlying network --- opens up an interesting application of Xen for testing/simulating wide-area deployments.

  To this end an existing system for allowing the simulation of physical networks: DummyNet, is a perfect match to be incorporated with Xen. It then becomes feasible to construct a micro-planetlab: a network of hosts with Internet-wide properties (latency, delay etc)

  It is anticipated that a DummyNet system would itself be a VM with virtualised network interfaces offered to other domains on the system; the DummyNet virtual machine can then define (and change) the parameters of each interface: bandwidth, latency to given destinations, etc.

  For the emulation of a network, a characterization of the properties of that network are essential. To this end, in addition to implementing DummyNet-type functionality within Xen, the student will need to understand it's limitations and perhaps, time-permitting, explore mechanisms to overcome those limitations.

  The link for Xen is above and this will find DummyNet.

  Note: this project requires a student that is comfortable with the C programming language.

• Virtual Discs
  The virtualization of operating systems provided by Xen allows us to intercept the interactions between an OS and the hardware devices that it uses. One example of this is block device interposition: messages to and from a physical disc can be redirected to a separate VM, where a virtual disc may be implemented.

  For this project, you will take advantage of existing code that allows interception of disc messages to build some examples of virtual discs. You will start with an (easy) initial example of building either an encrypted or compressed file store as a virtual disc. If this is successful, there will be plenty of opportunity to extend the functionality to include copy-on-write and "time-travel" discs.

  Thanks to Andrew Warfield for thinking of this one