These are not all of the things that I looked at when grading, but these are 
some of the most important points.

For the implementation part of the assignment:

- More marks were given for the baseline than for protocol 1 and protocol 2
  implementations. The marks for protocol 1 and protocol 2 were relatively
  easy to get if an attempt was made to implement the changes to the baseline
  case. Marks were primarily deducted for mistakes in the implementation of
  the baseline.

- Some marks were deducted if the code didn't use the independent random 
  number streams funcionality of CSim, for generating separate streams of 
  random numbers for each source of randomness in the simulation. Also, 
  marks were deducted for using multiple STREAMs, but forgetting to seed them
  using the stream_init function. This would cause all of your random number
  streams to have the same seed, and hence to generate the exact same numbers.

- All of the statistics should be collected and computed correctly for each 
  of the three experiments and a confidence interval computed. The confidence 
  interval had to use the correct formula and the correct t-value for 5 
  replications and a confidence level of 95%.

- Some common mistakes in computing the statistics include: giving the absolute 
  times for various states of the tug and the berths rather than proportions of 
  time, and summing the number of storms over all five simulations rather than 
  for each individual replication. While it is possible to see from the wording 
  of the question why some groups would have summed the number of storms over 
  the five replications, the whole point of doing multiple replications is to 
  obtain a point estimate with a confidence interval. Summing over the replications 
  is not as informative. 

- Another problem which cropped up had to do with computing the time average queue 
  length for the berthing and deberthing queues. The csim QTABLE has a function, 
  qtable_qlen(), which will compute this for you. The calculation involves taking 
  taking the average of the possible queue lengths weighted by the proportion of time 
  that the queue spent at each queue length. There is quite a bit of bookkeeping 
  involved in performing this computation and there isn't a simple formula which 
  will do it for you, which is why it was recommended that you use the QTABLE data 
  structure to take care of this.

- The tanker residence times consist of the total time spent by the tanker from the 
  time it arrives in the harbour until it is dropped off in the harbour by the tug. 
  This is the sum of the loading time, the berthing and deberthing time, and the
  time spent by the tankers waiting in queues. If your computed residence times were
  less than the mean loading time + 2 hours for berthing and deberthing then I docked
  marks, as this should have been a clear indication of something wrong either with
  your simulator or with the collection of the statistics. 

- The generated output should should have been close to that generated by the 
  solution implementation. Although, some variation in the final results was accepted,
  because it is a random experiment.

- Even if the code and the output submitted in the hard copy looked correct, 
  some marks were docked if the electronically submitted code did not compile 
  and run correctly. This was sufficiently emphasized in the hand-in 
  requirements.


For the report part of the assignment:

- The main thing that was expected in the report part of the assignment was
  some comments on the relative goodness of the three policy alternatives,
  with reference to the simulation results *and* the confidence intervals.

- There was very little difference in performance between the three policy
  alternatives, and the confidence intervals overlapped quite a bit. When
  making comparisons between the three protocols it was important to mention
  that even if one protocol had better performance than another that the
  differences were statistically insignificant.

- Some people asked in the newsgroup whether they should do the alternative
  design comparison from Sections 13.1 and 13.2 of the text. Although the 
  assignment sheet did not specifically ask for it, you were encouraged to 
  do it if you felt that was appropriate. Many groups did the analysis and 
  some marks were allocated for it. It was particularly useful in this 
  experiment because there was so much overlap between the three policy
  alternatives, that it was helpful to be able to show that the confidence 
  interval for the mean difference of many of the performance measures 
  contained zero.

- It was hoped that upon realizing that the number replications caused the 
  confidence intervals to be too wide to draw any strong conclusions, that
  you would either perform more replications or recommend that more replications 
  be performed to reduce the confidence intervals. However,almost no groups 
  did this, and no marks were allocated for it.


If you have any compliments or complaints about how your assignment was marked,
they should be directed to Dr. Graham first. Good luck on your final exams
and have a good summer!

P.S. Some people seemed confused about the difference between berthing a tanker 
and birthing a tanker. There's a big difference. Refer to your favourite 
dictionary.  :)