forum.alglib.net

What are the practical maximums for Linear Programming problems:
1) total variables
2) total constraints.

Is there any performance data available for commercial version for larger Linear Programming problems?

Hi!

There is no clearly specified maximum, the answer depends on the properties of the constraint matrix. With highly sparse constraint matrices (just a few nonzeros per row) you can have hundreds of thousands of variables with reasonable performance. Dense constraint matrices slow down algorithm and increase memory consumption, obviously. Algorithm won't fail as long as it has enough memory, though.

Presently commercial and free versions of LP solver have similar performance. We released LP solver as soon as it become stable enough for production use. Future versions may introduce some performance-related commercial-only improvements, but as for now - commercial and free editions are 100% same.

Can you describe your task in more details - variable count, constraint count, sparsity pattern of the constraint matrix?

I don't have all the size numbers yet.
The basic model is a financial retirement portfolio (savings, 401K, stocks, bonds, etc) with a 30 to 40 year planning horizon.
Hope to have size data soon - maybe next week, or week after.
Current model is in Excel and is just to limiting according to the portfolio manager.
Many kind thanks for your prompt reply.
Your guidance is quite helpful.
Will be in contact as soon as I know more.

Just to be sure: do you have binary 0-1 constraints, or integrality constraints? Present version of LP solver does not support MILP, only continuous LP is supported.

I think continuous LP is OK. At least to start with.
I'm still waiting for more details - a meeting with the principals is being organized for next week - they are all on travel now.