It Is Too Slow
JeanFrancoisPuget 2700028FGP Comments (2) Visits (11241)
"It is taking too long". Who didn't heard it at least once? I mean, who among those who thought they had successfully solved a challenging optimization problem?
This came again during a discussion with colleagues of mine in IBM consulting arm last week.
They have an unhappy customer.
This happens, why blog about it?
Well, the reason why the customer team is unhappy is worth telling. They are unhappy because they cannot scale the optimization solution. They can successfully solve a problem with 10,000 items (that was the initial request). They can even solve a much larger problem with 30,000 items. But now they want to solve a problem with 60,000 items. And it is taking too long, say 10 hours, when they would like it to run in 3.
Of course, we'll do what we need to do to improve efficiency, Fortunately there are several ways to do it. I may blog about these some time. But the point I would like to make today is different.
Customers will never be happy with optimization running times.
Because in fact they are happy.
No, I am not drunk. Not really I mean. Bear with me and revisit your judgement in few minutes please.
The reason they are unhappy is that they cannot scale to more complex problems. OK, you figured out this yourself. But why is it they want to scale much farther than their original request? Because optimization results are great; much better than they though. They now want to apply the magic stick to all of their problem! Hence the scalability issue.
I first heard about this 20 years ago. We were asked to solve a daily locomotive scheduling problem within a minute. We managed to solve it in one second. We thought the customer would be delighted.
No. The customer immediately ran the one week problem instead of the one day problem. It was taking like 10 minutes. And they asked us to solve it in one minute... I predicted that if we were up to the new challenge then they'd move to the monthly problem, asking for a solution within a minute.
Guess what? It happened exactly as predicted.
So, next time someone complains about running time, hear the true complaint: "I can't wait for optimization results, they are so useful".