Open Hypermedia Systems Working Group

Beans, RMI, Voyager

Recommendations

Whereas I am a bit sceptic with regards to Java RMI, I fully support Voyager. It is actually in my opinion unfair to compare RMI and Voyager as they are in different weight classes. RMI is bluntly speaking RPC for Java, whereas Voyager is a big and rich ORB architecture with elegant support for many things, RMI is nowhere close to achieving.

A very nice thing about Voyager is the fact that it does not belong in neither the DCOM nor the Corba camp. If ObjectSpace is able to fulfil their promises with regards to DCOM and Corba support, the Java programmers among us have a winner on their hands.

A drawback is that full Corba support is not yet here and DCOM is to arrive in mid 1998.

Turning to the points raised by Ken Anderson, does Voyager allow us to

• specify a set of related components.
• develop and run the components cheaply (e.g. for free) and on multiple platforms.
• exist in an Internet environment (e.g. with associated scale and security issues)

Voyager provides the tools to make all Java components(/classes) related and this combined with the ability to use IDL and integrate with Corba would seem to fulfil the first requirement.

Voyager is free for use in the binary form, and as such probably one the cheapest ORBs around. If we however require source code, we will have to pay ObjectSpace. As for multiple platforms - Voyager is 100% Java and should as such run on any Java Virtual Machine. While Voyager is Java-based, it is less Java-centric than for instance RMI, as the support for IDL and IIOP allows for integration with more than other Java programs.

Of the various ORB-systems I have read about Voyager is the most advanced with regards to networking. By letting lightweight agents handle messages, Voyager allows for very sophisticated future expansion. Voyager scales very well with regards to multicast messages, as it introduces the idea of Spaces and Subspaces of hosts - a multicast message is sent once to a Space, and the members of the Space then distributes the message among themselves. It is however a little early to decide whether Voyager is able to scale well in a large setting. When it comes to security, Voyager has the characteristics as the rest of the Java platform with security managers and sandboxes, plus the ability add more secure transportation layers.

What then are the weaknesses of Voyager? It relies on one company, it does not (yet) have all the features nor the speed, one might want, and the source code is not freely available. ObjectSpace is certainly not as large as Microsoft, but they are well respected for their container library (JGL), and have so far been able to hold their deadlines. Whether or not the free availability of the source code is a problem is a problem depends on what modifications we need (should we decide to go for Voyager) and on how ObjectSpace plan to fulfil their obligations.

It is my opinion that Voyager is the best ORB architecture available for Java today. It is very sophisticated in its own right and it supports Corba (and hopefully soon DCOM). A very nice aspect is that we need not abandon IDL as specification language, and given this, I find it highly likely that some of the early implementations of OHP will be running on Voyager servers.

Please send feedback to Niels Olof Bouvin. Last modified: Mon Dec 15 07:09:34 1997