Does 'just in time' = 'better late than never'?

The World-Wide Web is emerging as a medium for distributing platform-independent, intermediate-form programs. Most Java vendors have recently announced plans to construct 'just-in-time' systems, which translate the intermediate text into native code on demand. In this paper, we present experiments that show the benefits of just-in-time systems as compared with the traditional (compile prior to execution) systems. We introduce a new method - the continuous compiler - that can outperform just-in-time systems by overlapping compilation with program interpretation and native execution. Based on those results, we then present a smart just-in-time system that blends interpretation with native-code execution, thereby obtaining improved performance.