6.033 - Computer System Engineering (Spring 2004)

Design Project 2 - FAQ

posted 1:58 pm, May 5, 2004

Q>: (Note that this is an old question with a clarified answer). Can we assume that there is a maximum probability of sensor failure?

A: This should have read: you may assume that sensors fail substantially less than 50% of the time. If your analysis requires some maximum failure rate, we suggest you pick a number that seems reasonable (say 10% or 20%) and work with that. Be sure to state your assumption.

posted 10:50 pm, May 4, 2004

Q: I've been agonizing over a horrible probability problem in an effort to make experiments succeed more than 95% of the time. I just can't get the math to work out! What's going on here?

A: This probability analysis turns out to be much harder than we had intended. Rather than focusing on solving the problem with the particular parameters given in the design project, we suggest that you propose a technique that will reduce the probability of failure given that some sensors fail more often than 5% of the time, and give a quantitative and/or qualitative argument that demonstrates the effectiveness of your approach. Your primary objective should be to demonstrate that you have an intuition about how to solve this particular type of fault tolerance problem. If you already have written up a solution to the problem as stated, it's fine to leave it as-is, but make sure that you have given an intuitive justification for your solution.

posted 7:54 pm, May 3, 2004

Q: Can we assume that there is a maximum probability of sensor failure?

A: You may assume that sensors fail less than 50% of the time.

Q: I noticed that Table 1 lists a "prob(sensor fault)". What does this number represent? How does it relate to the per sensor/experiment failure rate referred to in Section 3.2?

A: This is a bug. You should assume that failure rates are specified per sensor, as described in Section 3.2.

posted 1:31 pm, April 25, 2004

Q: The design project simply states that the MTTF of rovers is 100 days. Is this failure process memoryless?

A: You may assume memoryless failures, or any other reasonable and well justified failure model. Be sure to state your assumption!

posted 8:15pm, April 13, 2004

Q: This question appeared previously. Note the new answer! What if the rover dug a hole and failed? If the coords aren't that accurate how does the replacement rover know to go to that hole and take a sample?

A: In general, you should assume that all experiments are both idempotent and atomic (look up these terms in the Chapter 8 Glossary if you are unsure of their meaning.) Thus, for the specific case of a drilling experiment, you may assume the rover can find and reuse the old hole, if the failed rover managed to offload information about the location of the hole before failing. If it didn't, the new rover must drill a new hole.

posted 3:00am, April 12, 2004

Q: Within a mission, can experiments can executed in any order? How do we tell which experiments depend on which other experiments?

A: Within a mission, experiments must be executed in exactly the order specified, regardless of clustering. (If NASA had considered the clusters of experiments to be independent, they would have issued them as separate missions.) Every experiment depends on the results of all previous experiments in that mission. The experiments in a mission need not be executed by a single rover, but the results of experiment N must be known by a rover to perform experiment N+1. For example, if experiment N was to drill a hole and N+1 is to take a soil sample from the hole, the rover which will execute N+1 must get the exact location, shape, and depth of the hole from the rover which executed N. You may assume that only the results of the immediately preceeding experiment are necessary to perform the next experiment.

posted 4:01pm, April 11, 2004

Q: Can the control center send data to rovers if they are at the edge of the crater? Rovers can only send data 100 meters; does the control center have the same limitation?

A: The control center's radio also has a 100 meter range.

posted 8:04pm, April 10, 2004

Q: Does NASA account for multiple trials of experiments in the their projected time?

A: No; the time is for a single run of the experiment.


Q: Do all experiments have to be run in order for scattered-cluster and scattered missions?



Q: The intro and result sections differ in describing the last time an ammendment can be made. Intro says when its been uploaded to base, results says when its been downloaded by NASA. which should we use?

A: The commit point is when the results reach the basestation, but amendments can arrive later than that. You should read about commit points in the notes and decide what to do.


Q: If Control Center tells two rovers to do the same mission, do the rovers know not to crush each other? can they both run the experiments at the same time?

A: The coords aren't that accurate so two rovers can run the same experiment at the same time.

Q: Is it ok to say: I'm going to use 802.11b and TCP for my communications protocol?

A: Yes, if you can explain to us how it solves the problems you care about.