Starting values of the variables are important to the success of the optimization. The closer the starting values of the variables are to the final and optimal values, the less time it takes to make the optimization run. If you choose the initial values to satisfy the constraints, Embed can skip the initial phase of finding variable settings that satisfy the constraint (that is, settings that are feasible).
When Embed begins with infeasible starting values, it must use a Phase I optimization process. This means that Embed uses the sum of the constraint violations as the objective function, temporarily ignoring the real objective function.
It is possible for Embed to terminate at the end of Phase I without finding a feasible starting point. If you believe that the problem can be made feasible, try setting different starting values for the variables and try again.