Papers on Smart Pointers in C++ and Evolutionary Delivery and a more detailed consideration of idiomatic use of the STL are also available at this site. For more information on the author, please take a look at my software and home pages.

Experiences with ORBs

Report on a Birds of a Feather session at Object Technology 97
Oxford, England, 15 April 1997 David Harvey

This BOF session brought together nearly thirty people to share their experiences of building distributed systems with ORB technology. A third of the group had implemented one or more ORB-based systems, another third were involved in evaluating the technology, the remainder came to see what was going on in the field. Here are some conclusions and observations.

Products

Unsurprisingly, the most widespread product reported in evaluation and use was Iona Technologies' Orbix. Other vendors represented (generally as singletons) were ICL (DAIS), ParcPlace (Distributed Smalltalk) and Visigenic (VisiBroker).

Criteria for evaluation

These included

support for COSS services
scalability
interoperability

A significant problem with evaluation was the lack of reference sites with significantly-sized implementations ('large' appears to mean 10-20 clients in many cases). I was surprised at the number of single-vendor evaluations in progress (to me, these are investigations, not evaluations).

Uses

The technology is being evaluated and used for a variety of systems, including

Data warehousing
Telecommunications
Live data-feed processing
Financial systems
Integration of legacy systems

Motivations for using the technology (beyond basic distribution support) included location transparency (deferring the decision on whether an object should reside in a client or server process) and integration with underlying ODB products.

Observations

Management distrust the technology.
For small-scale relatively simple architectures, a couple of individuals reported that 'out of the box' usage was considerably simpler than expected.

Hints

Limit ambition
Freeze interfaces
Keep interfaces simple (the narrower the better)
Take great care with transactional issues. Restrict resource locking to the duration of a single call at a server.

Problems

Quality and stability of implementations (particularly with respect to memory management)
Performance (marshalling of complex data caused one distributed system to become effectively CPU-bound)
Security (lack of support)
Scalability (lack of current support for transactions, no built-in load-balancing, or indeed tools to enable load to be determined)
Interoperability (between C++ and Java versions of ORBs, and between different vendors' products)
Lack of tools to assist in deployment and management of ORB-based systems
Recovery behaviour

Conclusion

At a final show of hands, the response to the question "Is this technology currently stable enough to support large-scale distribution" was hardly encouraging to the developers of this technology: only four of the group answered yes, and in all cases with reservations.

Maybe next year we'll have a more optimistic picture to present. Many thanks to all who attended!