Добавлено: Вс, 11 Янв 2004, 21:35 Заголовок сообщения: Интересная информация по оптимизации в MS Project
Дoвoльнo интpecныe oтвeты в фopyмe cпeциaлиcтoв Arinte, кoтopыe paзpaбoтaли cвoю cиcтeмy oптимизaции гpaфикoв paбoт для MS Project.
Thank you for your analysis. Your information is very interesting and useful. Below you find answers to your questions.
VNI> Test #1.
VNI> Resource Pool.
VNI> 10 projects.
VNI> Master project.
VNI> Level Better cannot work with this structure of projects
VNI> (overlocations is present). How Level Better work with resource
VNI> pool of MS Project?
As for resource pools and linked projects Level Better! here works just like Microsoft Project does. It levels all open linked projects.
Actually Level Better! cannot leave any overallocations in the solution. If it reports that solution found this solution will never contain overallocations. However Microsoft Project sometimes show overallocations where they are not present. Probably you have met this particular situation. Try to use Resource usage view to check whether the overallocation actually exists.
In some cases, however, Level Better! fails to find solution at all. In this case clear indication is given, stating that solution was not found. In this case schedule is left intact and of course original overallocations are left as well. There are different reasons preventing Level Better! from finding solution:
- There is no solution at all (resources are not enough etc.)
- Resources cannot be leveled within given interval. Interval is bound
(in case of leveling from start) by project start in one hand and
by expected project finish that defined in Level Better! leveling
dialog. ('Project finish not later than' input fields)
Level Better! does not work fine in case when some of linked projects are leveled from start and others are from end. In this case projects with leveling direction different from one of the master project are not taken into account. This is the third probable explanation of the situation you reported.
If you case is not one of the above, than we need the complete projects set for deeper analysis. We will also need the projects set if you need further explanation on why solution cannot be found by Level Better!
VNI> Test #2. (see attach)
VNI> "Commertional Printing Planning"
VNI> Microsoft Leveling process this project in 5 seconds
VNI> Level Better in 65 seconds (more than 10 times slowly). Randomized mode
VNI> - 60 seconds.
VNI> Duration result of Microsoft Leveling and Level Better is equal.
First of all, Level Better! is not expected to be very fast. If one wants to save weeks of project time it could wait some minutes for better schedule. So performance is not the key priority in Level Better! design. However in the test given I haven't seen big difference in the performance if Microsoft Project and Level Better!
Please note, that Level Better! uses public COM interfaces provided by Microsoft project for data transfer to and from computational engine. These data transfers are pretty slow because of COM technology in one hand and because of great limitations of the interfaces itself in other hand. The latter means that Level Better! cannot use the same mechanisms as Microsoft Project internally does, it obliged to use narrow and cumbersome interface instead and it drastically impacts performance.
As for computations themselves, Level Better! finds the
first solution than in a second on our test Celeron 450 machine. For Microsoft Project it takes about 10 sec to find the same solution. All time since solution found Level Better! spend trying to improve this solution (and to prove its optimality in non-random mode).
Finally solution found in non-random mode in most cases is proved to be optimal and no better solution exists under all constraints given.
VNI> eLabor leveling in MS Project Professional - 1 second and 10%
VNI> shortest duration
As I said solution found in non-random mode is almost always proved to be optimal under all constraints given. The most probable explanation of shorter schedule found by eLabor leveling is that eLabor violates some constraints. E.g. it can redistribute resources among tasks or treat some constraints differently from Microsoft Project (we at ArintE tried to treat constraints just like MSP does even if it has not much sense). For further analysis we need to work with schedule created by eLabor. If you could send us one, we can give you more detailed answer on this issue.
However, in some cases Level Better! fails to find absolutely optimal solution in non-random mode. And the Printing project might be the case.
Since the project optimization task is proved to be hard all optimization systems utilize some proprietary methods, that cannot work fine on all of the tasks. So some system can be better on one task while other system will be better on another one. For precise comparison of the optimization systems one need to solve number of tasks (the more the better) and compare statistics. The best comparison will be achieved on set of tasks natural for the particular business.
Thank your for your interest to Level Better! solution.