Hi folks,
Based on competitor experiences, below are a few notes for using the EC2 for the final competition evaluation.
If anyone has additional tips, please feel free to post them to the IPPC list.
Cheers,
Scott
(1) Time to copy over final competition instances
The archive of final competition instances is about 30 Mb, so it will take some time to copy over to the EC2. I recommend using the "wget" command to download the link I send directly to your EC2 node. I will email you the final competition instance link about 30-60 minutes before the start of your competition time slot.
(2) Beware of large instance memory limits on the EC2
It has been reported that if a planner exceeds the 7.5Gb memory limit of the EC2 large instance (which you should be using for the final competition), then the EC2 appears simply to kill the process rather than letting the memory overflow onto disk as virtual memory. This behavior cannot be 100% confirmed, but it appeared to happen in a recent evaluation run, so please be aware of this potential EC2 behavior.
Consequently, *please* test your planner out ahead of time on the *largest* instances from the test competition to see if this issue affects your planner.
(3) Dealing with temporary network outages
Network outages may disconnect your ssh session from the EC2, don't let this kill your planner. Instead use a command like nohup that will keep your planning process running even if you log off (intentionally or unintentionally). For example, to run the Servers, I use the following command:
nohup ./run rddl.competition.Server files/test_comp/rddl/ 2300 > output.txt &
Then, if you want to keep track of the 30 lines output by your planner, just use
tail -30 output.txt
If you need to kill your planner and restart, use
ps -ax | grep "[your planner's command line executable name]"
to find the process ID, then kill it using
kill -9 [your planner's process-id]