Kernel died, restarting for big size problem

[Siamak Naderi]

Hello, 

I am using Gurobi 8.0.0 and Python 3.6 in Windows 7. I have a code in which in the first step I solve MILP with a Lagrangian relaxation based heuristic and in the second step I solve the same problem with a time limit with is equal to execution time of LR heuristic. I face with no problem in the first step however, in the second step I receive "Kernel died, restarting". Moreover, for problems with few number of decision variables and constraints, I don't have this problem while for large scale problems I do. 

What causes this? 

Any help is appreciated, 

Siamak   

[Tobias Achterberg]

May it be that you run out of memory? You could watch the task manager of Windows.

Regards,

Tobias

[Siamak Naderi]

Dear Tobias, 

It seems that problem is with memory. How can I possibly use both RAM and Hard disk while solving a problem? Is there a way to modify the settings in this way?

Bests, 

Siamak

[Siamak Naderi]

Dear Tobias,

How can I switch to hard disk to avoid this error? I want to use both memory and hard disk. 

Bests, 

Siamak 

On Wednesday, June 27, 2018 at 12:23:42 AM UTC+3, Tobias Achterberg wrote:

[Fillipe Feitosa]

I got this problem because of Memory.

Using Linux (deepin) with python 3 and jupyter.

The solution was to give a breath to the O.S with 18GB of swap memory. 8Gb of Ram was not enough.

[Tobias Achterberg]

There are two ways to use the hard disk for additional memory in Gurobi.

One is the general purpose swap space supported by the operating system. This you need to
enable in your operating system, it has nothing to do with Gurobi.

The other one is for storing nodes of a MIP search tree to disk. If your memory issue
comes from very large trees, then this can be useful. Because Gurobi knows which of the
nodes should be stored to disk and which should be kept in memory, and how to organize
best the nodes in the node files, this is much more efficient than the general purpose
swap space of the operating system. But as I said, this is only applicable for large
search trees. You activate this feature using the "NodefileStart" parameter, see
http://www.gurobi.com/documentation/8.0/refman/nodefilestart.html#parameter:NodefileStart


Best regards,

Tobias