This morning I submitted an AMPL model for an LP to be solved by GUROBI. The problem has hundreds of thousands of variables and takes hours to solve locally on my machine (using LOQO). As I had already solved the problem on my machine, my main interest was to get a sense of the cpu time to solve this problem on the NEOS server. Hence, the "AMPL commands" file contained just one command: "solve". I did not ask for any output. I figured NEOS would then just send me the optimal objective function value and perhaps (hopefully) the cpu time required to solve the problem (and maybe an iteration log). I was very impressed to discover that the whole thing took only about a minute. I got an email telling me the job was done but it didn't tell me the optimal objective function value nor did it tell me the cpu time (which must have been very quick!). Being a little bit suspicious, I resubmitted the problem with one extra line added to the "AMPL commands" file. The extra line just asked that the optimal objective function value be displayed. I submitted this job early this morning. Apparently, it is still running (about 8 hours so far). I now think that maybe GUROBI and LOQO are actually similar in computational efficiency and I suspect that the quick turnarounds this morning were the result of NEOS not doing anything because I had not explicitly asked for any output. Is this correct thinking on my part?
--Bob
Robert Vanderbei
"Just what the truth is, I can't say anymore." -- Moody Blues
