Dear experts, I am a Ph.D. student centers in logistic distribution. Now I am using CPLEX to solve a complex linear combinational optimization problem. But even small scale problems run on a 4GB RAM computer, I am ofen faced up with the "Out of Memory" error. I am confused. I have tried many methods, such as tune the parameters, but this error still happens. I want to know is there other setting I could adopt? Or other problem I should pay attention? Now I see that a new solver GUROBI, which seems to have higher efficience than CPLEX ? I want to know in GUROBI, can I use column-generation or decomposition algorithm as in CPLEX? And in GUROBI, can the "Out of Memory" error happen? Thank you very much. Since this error has bothered me for a long time, I want to have a better solution. Best Regards!

asked 22 Jun '13, 10:20

LinYuan's gravatar image

LinYuan
80617
accept rate: 0%

edited 26 Jul '13, 12:19

fbahr's gravatar image

fbahr ♦
4.6k716


From IBM support:

From CPLEX (v12.5) documentation:

And: Yes, Gurobi can run into "out of memory" situations, too.

link

answered 22 Jun '13, 11:02

fbahr's gravatar image

fbahr ♦
4.6k716
accept rate: 13%

Dear Bahr, thank you very much! I can always get very useful instructions from you help. And since I am a beginner of Gurobi, in this solver can I design my own branch-and-bound-and-price algorithm and column-generation algorithm, based on my characteristics problem ?

(26 Jun '13, 10:34) LinYuan

No - or, as it seems, it's at least not a trivial task:

Can you modify the branch-and-bound algorithm or create a branch-cut-and-price algorithm?

The only way to create a custom algorithm like this is to stop solving, modify the model, and start solving again.

Callbacks provide a - afaik - very basic mean to inspect and affect the solver's progress, but you can't just simply modify the branching behavior of the built-in implementation or hook your own one in.

(26 Jun '13, 10:57) fbahr ♦

For more/better advice, check Gurobi's Google group.

(26 Jun '13, 10:57) fbahr ♦

Try reducing the number of threads you're using.

link

answered 23 Jun '13, 04:54

jfpuget's gravatar image

jfpuget
2.5k310
accept rate: 8%

Thank you very much. In my test, I find that the parameter tuning may be help. I also want to know that in GUROBI, can I design my own branch-and-bound-and-price algorithm and column-generation algorithm as in CPLEX, based on my characteristics problem ?

(26 Jun '13, 10:36) LinYuan

I'm definitely not the person to ask for gurobi help ;)

(27 Jun '13, 14:39) jfpuget
1

They have an active discussion group here : https://groups.google.com/forum/#!forum/gurobi

(27 Jun '13, 14:44) Bo Jensen ♦

Here's one parameter that you should consider looking: node file storage mode. mip strategy file (as named in CPLEX).

Essentially, this means that the solver stores the nodefile to the hard disk when it reaches a certain size. Although it may slow down the execution of the algorithm, it is an acceptable compromise when solving large models that do not require fast solving.

In Gurobi the switch is called NodefileStart parameter, while it's called mip.strategy.file in CPLEX.

link

answered 04 Jul '13, 01:35

ORNinja's gravatar image

ORNinja
1311
accept rate: 0%

Your answer
toggle preview

Follow this question

By Email:

Once you sign in you will be able to subscribe for any updates here

By RSS:

Answers

Answers and Comments

Markdown Basics

  • *italic* or _italic_
  • **bold** or __bold__
  • link:[text](http://url.com/ "Title")
  • image?![alt text](/path/img.jpg "Title")
  • numbered list: 1. Foo 2. Bar
  • to add a line break simply add two spaces to where you would like the new line to be.
  • basic HTML tags are also supported

Tags:

×191
×22
×5
×4

Asked: 22 Jun '13, 10:20

Seen: 4,925 times

Last updated: 26 Jul '13, 12:19

OR-Exchange! Your site for questions, answers, and announcements about operations research.