Resources for thesis writing
From Computing Project Repository
General structure
A report has a general structure:
- Introduction
- Motivation and aims
- Literature survey of related work
- Methodology
- Results and their analysis
- Conclusions and suggestions for further work
Your references to literature should not only be in an isolated "literature survey". In general, your report should have an introduction and conclusion that is easily understood by a non-expert, and a technical part that is detailed enough so that another expert could reproduce your results. Similarly your discussion of the literature can be at a high level in the introduction and conclusions, and at a more detailed technical level in the technical part.
Experimental methods
Control experiments
A control experiment means to take two identical experimental setups, under identical environmental conditions, change one of them and measure the effect of that change. One of the systems remains unchanged for comparison.
You should mention which environmental factors (factors that you do not control yourself) which can change, and draw them as a cause tree. See Analytical Network and System Administration (http://www.iu.hio.no/SystemAdmin/analyticalbook.html).
- Comparing the same system at different times is not the same
- Something might have changed in the environment since the last time it was measured.
- Need to think about the effect of making multiple measurements on both systems at the same time. When? How?
Checklists for discussing performance measurements
- Remember that all measurements are uncertain. Draw a cause/fault tree that shows the different likely sources of uncertainty. Can you estimate how large each source is?
- What controls will you use to understand your results?
- Remember to `de-trend' your measurements, e.g. remove periodic variations in the average.
- Remember to repeat measurements and measure the distribution of values about the mean. Is it symmetrical, asymmetrical? Why?
- What is the flow diagram for activity in your system?
- Where are the sources and the sinks of activity? (Cause and effect)
- What are the overheads versus the processing times? (How does this change under stress testing?)
- When several sources come together, how will you demonstrate the effect of bottlenecks in performance?
- When there are random arrival events involved in a system:
- What is the effect of short bursts of many events, or long periods of calm? (Burstiness)
- Plot the distribution of inter-arrival times. Is it Poisson, power law etc.? What does this tell you?
- If several random processes flow together in a confluence, what is the behaviour of the combined stream?