How to solve 3D multi bin-packing with OpenOpt

Question

I'm learning about optimization and I'm new to OpenOpt.

I would like to represent processes where each have 3 resources usage indicators (CPU, memory and network), and I would like to assign N processes to groups / bins according to the following restrictions:

sum(cpu) within a group < 100
sum(mem) within a group < 100
sum(net) within a group < 100
Minimize(number of groups) or maximize the sum of each resource within a group.

Ideally I would like to have this type of output:

VM 1 assigned to group 1
VM 2 assigned to group 1
VM 3 assigned to group 1
VM 4 assigned to group 2
VM 5 assigned to group 2
VM 6 assigned to group 3
... and so on

Quesiton: How can I do that? If it is not possible to do this with OpenOpt, is there any other lib that could help me with this?

Here my initial code: https://github.com/vonpupp/mdbp/blob/master/ksp_2.py

Many thanks!

Answer 1

The idea is to assign a weight on each item (in this case, all are the same), and maximize the number of items limited by the constraints.

This is how the items are created:

def gen_vms():
    tg = tracegen.TraceGenerator()
    trace = tg.gen_trace()
    items = [{
                  'name': 'item %d' % i,
                  'weight': 1,
                  'cpu': t[0],
                  'mem': t[1],
                  'disk': t[2],
                  'net': t[3],
                  'n': 1
             } for i,t in islice(enumerate(trace), 200)]
    return items

And this is how the constraints are created:

def add_constraint(values, constraint):
    return values[constraint] < 99

def add_constraints(values, constraint_list):
    return [add_constraint(values, constraint) for constraint in constraint_list]

def gen_costraints(constraint_list):
    global constraints
    constraints = lambda values: (add_constraints(values, constraint_list))

The solving process is like this:

def solve_host():
    global items
    global constraints
    p = KSP('weight', list(items), constraints = constraints)
    result = p.solve('glpk', iprint = -1)

Further details on how to do this can be found here: https://github.com/vonpupp/2013-sbrc-experiments/blob/e2e8a2be320c8f77d67a5bc6bb822510564e80f3/myksp_oo.py