[NSWI131] Performance grading

[Petr Tůma]

Hi Tekla,

> You said that we need 80 points from assignments to pass the course.
> Does that mean that the exam project is not mandatory? If it is
> mandatory can you please provide some guidelines on what should it
> contain or what to pay attention to?

Sorry about the confusion, let me clarify. There are two milestones to completing the course: the credit (zapocet), awarded for successful lab participation, and the exam, where you get the final grade. The 80 points limit is for the credit, you still need to do the exam project to complete the course.

About the exam project. The requirement is to do an "end-to-end" performance experiment:

- You should describe your goals, what it is you are trying to evaluate, (dis)prove, etc.

- You should describe how you designed your experiment, what data and how it will collect, and how it will help you meet the goal posed above.

- You should conduct your experiment to collect the measurements.

- You should analyze the collected measurements and make the corresponding conclusions.

- You should shrink wrap everything into a package where somebody else can reproduce your results, that is, see and use the design documents, the results report, and whatever code or scripts you used.

Once you have the package, just send it to me, I will take a look and possibly ask some questions, and that will be the basis for the final grade. To decide the grade, I will be looking at what the experiment goals were, whether the methods you have picked can help address the goals, whether you have analyzed the measurements and presented the conclusions clearly, and so on.

As for the exact topic, this is up to you - I do not want to hand out more work unrelated to whatever it is you are doing otherwise. Instead, I would hope that you are working on some other project - your thesis for example - where some sort of experimental evaluation could come in handy. So you can do that for this exam, and reuse it later. If you are at a loss and have no ideas for a project, I can of course generate some tips :-) about the size, what I have in mind should be a day or two of experimenting and writing a report, which should be a few pages but not much more than that. In the past years, people e.g. evaluated the use of various vectorizing instructions in matrix multiplication, measured the overhead of code isolation frameworks in terms of execution speed, and so on.

Hope this helps, do not hesitate to ask if anything remains unclear.

Petr