Project ideas
Revision as of 08:37, 5 Jun 2005; view current revision
←Older revision | Newer revision→
Open projects
Complexity of administration tasks
In this project you will describe measurements of task complexity, as seen from configuration languages and human processes. You can use the language of the theory of computation, process and fault theory, use concepts like cut-sets and flows (Petri nets) etc. to see how complexity can be measured and perhaps reduced.
Stability of routing (Uninett)
Uninett are building a redundant backbone with a regional ring structure. We have experienced that the routine configuration does not always do what we want. We would like to develop techniques for measuring the stability and availability of our routes. This can be done by observing the IS-IS routing protocol transactions at several places in the network and developing models for the stability. One possibility is to use Uninett's ping-traceroute tool mping which has 10 years of data for packages loss and delays from the backbone. http://drift.uninett.no/mping. (Olav.Kvittem@uninett.no)
BGP analysis (Caida)
TBA
Sampling of Netflow Uninett
With today's high speed networks, routers cannot manage to generate complete netflow data for every packet. They have to perform restricted sampling. This project will study the effect that different methods of sampling have on the precision of the Netflow data. In order to carry out the project, a day's worth of network traffic will be stored using a passive measuring device. From this, the Netflow data will be generated with and without sampling in order to compare the results. (Arne.Oslebo@uninett.no)
Redesign of TCP dump
The current output of TCP dump is poorly suited for machine analysis. It is difficult to parse because there is no consistent use of formatting and its output is designed to be read on the shell by a human. In this project, a new model for output will be developed based on a theoretical classification of the main protocols and exceptions. The source code of tcpdump and/or libpcap can then be modified to generate the output. Other interfaces can also be explored, i.e. direct pipelines to software like IDS or cfengine.
Infrastructure simplification (IBM++)
Today, large and small companies often have a computing infrastructure that has grown more organically than rationally. This results in unnecessary dependencies and over-complicated solutions. In this project, you will consider the infra-structure of two example companies, and use analytical techniques to compare their aims with their system infrastructure. This includes finding and eliminating resource bottlenecks and security vulnerabilities in their architectures and recommending meaurable improvements.
Supercomputer to workstation migration (IBM++)
Why are customers migrating away from supercomputers? How does this process happen? What are the rewards? Are there are rewards going the other way? How do we work out "Total cost of ownership" for a mainframe or workstation? The project is about collecting data from companies that have migrated, using a questionnaire and performance data. The project should try to explain the findings with an analytical model, in terms of real payoffs and perceived payoffs.
Supercomputer versus Unix clustering (IBM++)
Clustering of servers. This project will involve a comparison between the UNIX clustering implementations and the mainframe clustering solution (parallell sysplex). Two and three tier application solutions - a comparison involving WebSphere (or any other application server) in a two tier and three tier solution.
User survey of practices using cfengine
Cfengine is a configuration management tool that is widely used. This project considers the design of a questionnaire about user practices with cfengine and collecting data from users. Results should be classified and the results presented geographically and for large/small organizations etc in order to correlated practices. Do a Cavalli-Sforza kind of PCA analysis to find trends.
System administration practices
This project considers the design of a very ambitious questionnaire about system administration practices. It involves collecting data from users. Results should be classified and the results presented geographically and for large/small organizations etc in order to correlated practices. Do a Cavalli-Sforza kind of PCA analysis to find trends.
Simulation of peer to peer opportunistic backup
Keeping multiple copies of data in a distributed ad hoc system can lead to inconsistency of versions and, in the worst case, the loss of data. In this project we simulate mobile ad hoc networks where hosts use opportunistic copying to back a backup of their data. This could lead to multiple copies in different states of repair (different versions). If there is a sudden catastophe or mistake, what is the probability that they will be able to get their data back, in the right version? (Think of bit-torrent?)
Comparison of NFS, Sambda and AFS
There are many file systems in current use. This project is about comparing and testing the functionality, management and performance of these three filesystems in Windows and Unix networks.
Storage planning and management
There are many strategies for redundant storage (backup). What are the advantages of Storage Area Networks (SAN)? How do these compare to things like NFS and AFS? How does RAID's instant replication compare to the use of a daily backup?
Design of a User friendly VPN/WLAN solution for all platforms
Setting up wireless is not as easy as it should be. Analyse the causes of problems, especially on linux, and design a solution that is provably "secure" and suitably automated.
User job administration in clusters (TMJ)
The project consists of the contruction of a tool for automated job submission instead of operator controlled job submission. Analytical content?
Voluntary cooperation in routing and networking
Spread of peer to peer protocols and ad hoc networking.
Existing P2P protocols include BGP. Analysis of peering practices in Norway or Netherlands -- this requires access to the BGP peering tables of the NIX (Uninett/Surfnet).
Task scheduling
AI enhanced help-desk. Examine methods for effectively scheduling tasks based on a help-desk parameter model in a human-computer system. Can the system emply machine learning techniques to improve scheduling performance? (e.g. like Bayesian learning)
Configuration Analysis
Configuration management requires one to formulate a policy that describes the approximate state of a system. Policy determination can often take a long time to formulate and it is easier for administrators to customize a system and then to build a model based in it. In this project we create a tool that scan a workstation or computer system and turn it into a prototype cfagent configuration. The tool looks for possible contradictions in policy by looking for loops in a graph.
Cfenvd correlation analysis
The cfengine environment daemon collects network statistics from individual hosts, but it does not perform any cross correlation of the data and findings. In this project, we develop a data collection and correlation tool with frequent updates, where an administrator can obtain an overview of the data from a central location. The tool should also perform a Principal Component Analysis or correlation centrality analysis to find the most important sources and recipients of different kinds of resources in the network.
Virtual ad hoc networks
Ad hoc networks are increasingly important both in local areas, where like-minded users need to communicate temporarily, and also in a pervasive computing scenario. This project is about simulating ad hoc mobile communications in a three dimensional space using GNU/Linux virtual machines. The model should allow for terrain variations and obstacles that can alter the signal strength between the hosts. This will allow us to perform management studies and test routing protocols.
Cluster approach to spam filtering
Bayesian spam analysis and filtering. ISPs need methods of handling huge amounts of E-mail. How could we use cluster technology to make a cheap and extensible solution?
Cluster approach to packet analysis
Compare the performance of a cluster for analyzing packet scans. How do we split up the input stream amongst the nodes? Can we do machine learning in real time? Computation of eigenvalue centrality or principal components on a cluster.
Security access control models
Filesystem ACLs are complex but powerful, Unix permissions are simple but primitive. Can we evaluate these two models of and find a game strategy for the best and most efficient, easily admininistered model.
Compare the multitasking performance of server OS
We have observed that a slow solaris machine is generally better as a server than a faster Linux machine. Is this true? If so why? How does the server performance of NT, Solaris and Linux compare? Contact sun microsystems for hardware?
Dependency graph analysis
Modern operating systems have extremely complex layers of software that have multiple dependencies. In this project we examine the graph of software dependencies in common operating systems and see how vulnerable they are to failure or infection by viruses. By examining different versions of operating sytsems, can we answer the question: how is this complexity changing in different versions of operating systems? ie. are operating systems getting more or less complex?
Intrusion detection language
Most anomaly languages use regular expressions to find patterns in text. Are regular expressions sufficient? Can more general automata use used to match patterns more successfully? Is that what IDS systems actually do in practice? If Bayesian learning from regular expressions equivalent to a more general language? This is a theoretical project in which we build a model of a learning system that uses regular expressions to perform pattern recognition and compares it to more advanced language forms from the Chomsky hierarchy.
Theoretical configuration language
Look at cfengine configuration and find out what kind of computational language it generates in terms of operations on a system. Would a simple automation suffice to repair damage? Do we need a full Turing machine? Grammar of changes
Grace under pressure
This project is about user satisfaction. Stress symptoms - stress relief. How can we measure the level of stress on users and admins and see how many mistakes this results in? (Security) Similar studies have been made in Nuclear industries. If we believe that happy users are good users, how can we estimate the security risk associated with user stress in a system of a certain complexity?
Security in social networks
Analyze the probable security of a social network. Create an experiment by first constructing the graph of associations in a particular group, and then seeing how a secret spreads. We need to get someone to play along -- e.g. here in the college, we can collect data using a web form as a questionnaire asking: who are the people with whom you talk? How often do people actually check and maintain their file permissions? Look at models for epedemic spreading.
Computer assisted forensics
Knowledge based systems using social metrics to solve forensic problems. Applying graph theoretical methods to generalized problem solving. This could be a continuation of the Archipelago project.
Cfengine micro client for Ximian PDA/Mobile phone
Managing pervasive mobile devices. Map out the possible management and maintenance tasks of mobile devices, including automatic backup, garbage collection and security. Make a viability study of using cfengine on these devices, in a cut down form. Implement?
Large Scale Change management
The opposite side of the coin to configuration and maintenance is the management of significant change, e.g. upgrades, re-design and replacement. Can change be implemented without disruption to service? Does this idea contradict the idea of convergence referred to in section? Planning changes of infra-structure can be dealt with in two ways:
- Deconstruction followed by reconstruction.
- Change of policy description followed by convergence to a new state.
Server on demand for lab-education
Presenting students with a usable test-ground in sysdem administration is a challange in todays educational systems. What are the possibilities for an unsupervised web-based broker system that gives students quotas on servers that may be utilized for virtual machines. Students can have as many virtual machines as long as their total resource consumtion is under their quota. A web form is translated into a configuration and the virtual machine is buildt and started and is ready for the student.
Self configuring VPN's accross arbitrary networks
The goal is to design and implement a type of software that can auto configure and set up virtual private networks beetween arbitrary subnets. A negotiation and configuration protocol has to be developed. Can existing configuration management tools be used. This software can then run on a Linux-based bootCD and everywhere they are booted, they will detect and connect to the other running instances.
Benefits of virtual machines in sysadm egucation
How has the use of virtual machines improved the system administration education quality? What are the real challanges and benefits?
Network monitoring and visualization aimed at control-system design
Several systems have been designed in order to alert the system administrator and user when anomalies or pre-defined situations occur. How can network usage be translated into a usable representations that follow the guidelines of traditional control-system design? Are the classical web-based visualization front-ends (plots, lists, graphs,mail) usable compared to criteria given in control systems? What design-guidelines can be idetified for the developement of security/system/network monitoring systems?
Ideas already used
Traffic patterns with passive measurement (Uninett)
UNINETT user passive measurement with optical splitters and specialized measuring interfaces to trace traffic with gigabit speeds. We would like to investigate the structure and patterns in these data, It is of particular interest to classify the traffic belonging to different services and protocols. Intersting angles on this include file-sharing, attacks, and video streaming. The project is associated with: http://www.ist-scampi.org (Olav.Kvittem@uninett.no)
Quality of service in TCP transmissions (Uninett)
TCP is the dominant protocol for Internet services. We would like to analyse the quality of services used by users in TCP sessions, and define metrics for performance, package loss, resends and window sizes in the behaviour. The project will study TCP in detail and apply the results to the packet trace data obtained by passive measurement (see previous project). The project is associated with: http://www.ist-scampi.org (Olav.Kvittem@uninett.no)
Other ideas:
1. Real-time protocols (a survey)
What protocols are availabe, particularly for real-time VR (e.g.
VRTP)
2. Making a distributed database. How to distribute and store real-time
data. LDAP, XML, others?
3. Visualizing "any" data. Networks, buildings, network flow, anomalies
etc.
4. User interfaces with respect to the data visualized in the above
example.
5. Visualizing a MLN-network in 3D.
6. Simuating networks, buildings etc. to make analyzing and visualizing
e.g. network topologies more easy.