Executing an incoming job on the least loaded of the machines of a distributed system can considerably improve its response time. To make the right choice, it is useful to have some information both about the load on the machines of the system and about the job we are considering. The information about the load ought to indicate the resources a machine is able to give, whereas the information about the job must tell which resources the job is going to need. We developed a multivariable scheme to distribute load information and to match a machine's available resources with a job's specific requirements. The experiments we performed with a prototype implementation show that our tool is able to make the right choice on a set of test jobs between 55 and 88% of the times. Our purpose is to shed light on some controversial issues, in order to prune the intricate complexity of the problem and open the way to future more general implementations.