Summary of article: Len DiMaggio, "Orchestrating Integration Testing", STQE Magazine July/August 2003.

Systems are put together from components supplied by many parties. They may also be distributed.
The general question: Which variables in local systems can have which impact globally? Normally, they should have been identified in the design. But some are not. If the system is composed of many components, both COTS and specially developed ones, the integration test just has to proceed and variables identified during the test, from system logs. Run transactions through the system, check their results, and have all logging facilities on. The logs will contain information on.

Lists of places where subsystems expect to find executable programs or libaries (registry, for example)
Which database connections were established and closed by whom and when?
Which communication ports were assigned for what?
How were data converted and where?

Change the variables:

Exchange databases with other vendor's ones. Exchange versions. Exchange network support... See if something bad happens when doing legal operations of such kind.

Check failed transactions. How will they be logged, recovered, and is information to users and support staff good enough?
Check the logs for startup, shutdown and data processing. Is everything like it should be? What about cleanup?

Database compatibility

Maybe several databases, supplied by several vendors. Not all support all data types or all in the same way. DATE is especially bad.

Data may be converted. Trouble in doing so. Follow some sample data through the system. Check the stored values, no only the end result.
XML data parsing: MOST databases support XML data, some maybe not. Multiple parsers, each requiring different supporting software, even conflicting, may be in use. Parsers also contain problems. Send data in XML format through the parsers and check the result in the databases. Check also performance.
SQL syntax differences: For example reserved words. ("OID" is reserved in Oracle).

Inter program dependencies and start order

Some processes on one machine may depend on other processes on other ones. For example application server on database server. Could this go wrong? Naive users starting the whole thing up? Initializing processes taking too long time?

Is the required start order supportable? New processes coming in between the others mat have side effects to the ones coming after. Order may not even be documented.
Break the chain of dependencies in distributed systems. What happens if one process or server is down and comes up again? Are there still informative error messages reported to both users and logs? Are dead processes restarted by the monitoring process? Do processes recover from network problems (overload, outage, transmission errors)? Most important is to check what happens when database servers are down.

Installation and configuration

Directories, files and environment variable settings (including PC registry) common to multiple integrated subsystems should not be in conflict. (One subsystem requiring a differing value from another for the same variable).

Communication ports in conflict? Who uses which ports and when. Some conflicts not easy to find as not all ports may always be in use. Netstat command and etc/services file (PC and Unix) good to look into.
Requirement for different versions of supporting software
Is configuration data consistent over the systems? Read the configuration files.
Differing operating environment. A program may require certain levels of user privileges in conflict with other programs. What user accounts are created during installation and are their privileges correct?
Is the default configuration / installation supportable? Choosing default installation by users may result in inefficient or insecure installation. Check in BugNet, BugTraq and user group FAQs.<