An example of such a performance vs. simplicity tradeoff arises in the choice of a dispatching mechanism for the virtual instruction set. A simple virtual machine might dispatch bytecodes via a case statement, a more complex one might employ some form of threaded code, and a very high-performance one might translate bytecodes into native code dynamically. Each of these alternatives may have indirect complexity, portability, and performance costs. Case statements make hardware branch prediction difficult thus triggering pipeline stalls. Dynamic code generation requires an architecture-specific code generator and careful synchronization between the processor caches and the instruction fetch pipeline. Clearly, the case statement will be slower and the dynamic translator will be harder to port, but when performance and portability are both important, how does one choose? We'd like to understand tradeoffs between virtual machine design choices as well as we understand tradeoffs between different sorting algorithms.

While investigating such tradeoffs is one theme of this workshop, we encourage papers about every aspect of virtual machine implementation including:

• the impact of processor/memory architecture on performance
• history: what can we learn from the 1970's and 1980's
• using object-oriented technology to build virtual machines
• embedded virtual machines
• server virtual machines
• multi-language virtual machines

Since there have been a number of excellent OOSPLA workshops on garbage collection, we would like to gently encourage papers focusing on other aspects of virtual machine design.

Organizers:

John Maloney, Walt Disney Imagineering
Email: John.Maloney@disney.com

Antero Taivalsaari, Sun Microsystems Laboratories

Dan Ingalls, Walt Disney Imagineering