The process of defining requirements for "real-time Java," and subsequently the two competing specifications for it, began with considerable thought by the participants about what the term "real-time Java" could and should mean. Now two corresponding requirements and specification processes have begun for "distributed real-time Java."This paper summarizes some ideas about that term and an approach to an initial specification. The approach is based on providing a natural and minimal mechanistic extension to Remote Method Invocation (RMI) that facilitates real-time distributed computing in general and real-time distributed (in the sense of trans-node) programming in particular. Here, "real-time RMI" is not defined as "real fast" or even "real predictable" RMI (although those are both important attributes). Instead, it means that the timeliness properties of computational entities are preserved when the entities perform RMI's and RETURN's that span physical nodes. A similar approach has been proven effective in several other distributed real-time contexts, and is a primary feature of the unified proposal to OMG for dynamic real-time CORBA.