JuMP, excessive memory usage?

Specific Domains Optimization (Mathematical)
15

For posterity’s sake, I profiled memory allocation throughout my JuMP call. Here is the corresponding gist.

Summary of findings:

  • Data generation requires 15MB of memory allocation.
  • JuMP allocates ~600 MB at the solve() call
  • Another ~600 MB are allocated at the MathProgBase.initialize() call.
  • These two lines account for almost the entire 1.22GB that are allocated for the entire optimization.
  • When breaking down the MathProgBase.initialize() call, it appears that the second-order derivative computation is what is driving this, as it accounts for 79% of the allocated memory.

As Miles mentioned above, this doesn’t seem to be unintended behavior, but looks more like an issue with the second-order derivatives being too dense. This may also be what is driving @mauro’s problem, too.

1 Like