[Clp] Solving a model many times with small changes

[Drotos Marton]

Dear list members,

I am writing a Branch&Bound procedure with Langrangian relaxation (using an 
LP). This means that a CLPSimplex model is solved many times consecutively 
with small changes in the objective function / fixing some variables. After 
modifying the objective function and bounds on some columns I solve the model 
with ClpSimplex::dual(0, 7) (it seems that the dual is slightly faster for my 
problem than the primal). I suppose this way the simplex method uses as much 
information from the previous run as possible.

Are there any tricks or 'best practices' to accelerate this procedure (e.g. to 
use more information from previous solutions)?

The size of the model is around 5000-10000 cols x 500-1000 rows. Most of the 
colums are 0, the columns have bounds 0 <= x <= 1. All rows are either = 1 or 
<= 1, and the coeffient of the variables in the rows is 0 or 1. I use 
the 'official' Debian Lenny package of CLP.

Thank you for your help,

Marton Drotos