FU Berlin, Fachbereich Mathematik und Informatik, Institut für Informatik

Vortrag des Informatik-Kolloquiums

Although more than a hundred Concurrent Object-Oriented Languages (COOLs) have been introduced in the past, only slightly more than a third of them are targeted towards distributed environments. This is due to the fact that distributed environments do not only increase the amount of work language developers face when writing compilers and run-time systems but it is caused by the fact that object and process locality becomes the crucial performance issue that can no longer be ignored.

A major goal of our group in Karlsruhe is to contribute locality optimization techniques for COOLs. The first significant step in this direction has been done: We have a COOL and can compile it for heterogeneous networks of all relevant types of PCs and workstations.

The talk presents the results of the first step:

• We have designed a (very small) extension of Java for distributed programming with transparent object locality. The talk discusses the features of this language extension with respect to categories of existing COOLs.

• For this language we have implemented a compiler that produces ByteCode. For pure Java programs this compiler runs twice as fast as Sun's Java compiler and has thus been integrated into commercially available Java products. The talk presents the transformations to map our extension of Java to Java. It will be discussed how our main goal to address locality problems drove certain design decisions.

JavaParty is the platform for future work: The run-time system will be optimized based on forthcoming benchmark results. Locality optimization mechanisms will be studied.

More information can be found at http://wwwipd.ira.uka.de/JavaParty