From: neshnash_at_u.washington.edu
Date: Wed Feb 04 2004 - 18:12:16 PST
This paper is an interesting implementation of distributing system. It is an other eye opener that make us reflex about the dominance of current systems and question the need of alternatives.
The paper presents Emerald, an OO language designed to support fine-grained objects mobility. It starts with describing the advantages of migration and mobility and presents the new language Emerald as a solution that will enable relatively good performance by introducing language semantics targeting this problem and coupling the compiler and the runtime. The inefficiency of existing OO languages was described as the limitation of the types of objects to be created. The new language has a more powerful semantics that is oriented to building distributed systems. The performance is comparable to existing systems.
The paper describes the internal structure objects and communication methods. The examples were very efficient to get the ideas cross. User has the flexibility to control the location of the objects and the compiler is optimized to take the "best" migration decisions. Though, I have two concerns:
- Compiler decisions are made at compile time, it would rather be beneficial to take decisions at run-time based on network state, etc. Is it possible to implement a heuristic learning system that takes a better decisions as it learns about the environment it is working on?
- It seems that a copy of the object is kept on the node after the move operation. In addition to its performance hit (discussed on the paper), I can imagine a memory limitation problem. Copies of same objects on different nodes. How about multiple copies of same object in the same node?
This archive was generated by hypermail 2.1.6 : Wed Feb 04 2004 - 18:12:18 PST