I have not recently used CPLEX for large-scale problems, but about five years ago I used it for a very large-scale problem. That problem was too complex. When I would read it and do the solving, I could not get the feasible solution for the long period. 

I think that is okay, too. I don't have any complaints about that.

No issues.

I sometimes have an issue with the scalability but until now it has been reasonable. 

I haven't tried really, really big problems, so for your everyday problem it's fine.

But I did have some problems with error messages where, for example, it's infeasible, but instead of telling you it's infeasible it will give you something else.

If there is a lot of data, I have a hard time solving it with CPLEX.

The algorithms that I design are based on the premise that CPLEX struggles when there is a lot of data. So CPLEX on its own won't scale to the size of problems that I am trying to solve, though I kind of like this because it gives me something to do. But it seems that it can solve bigger and bigger problems every year.

The scalability really depends on the type of problems. For my particular types of problems, it's not as scalable because there are a large number of constraints to the problems. So, if I increase the number of variables, I significantly increase the number of constraints. For my particular problem, it's not as scalable. But that is the reason why I use these nuances for my problems.

CPLEX is limited to the linear programming and mixed integer programming. It is not extended to nonlinear programming, so I think the scalability can be an issue in that area. Though, most of the programs in the current industry are having linear programming problems, so I think CPLEX has a good future.

I find scalability to be more of an issue that comes from the modeling. If it doesn't scale, then it is usually the problem of the modeler. Some problems are just too big to give as a whole piece to solve and expect a reasonable answer in a reasonable amount of time.

I didn't implement any complex models in it, but I feel if I were to do that, I would use CPLEX for sure.