As I mentioned in the previous section about application-focused solutions, seriously consider finding creative ways to implement Oracle's global temporary tables. If the redo is associated with a restartable batch process, there is a good possibility of success. Using global temporary tables can cut redo generation by 50%, substantially reducing Oracle's overall IO requirements. This alone could solve the log file parallel write problem.

From a more Oracle internals perspective, delaying log writer background process writes so each write will be a full 1MB IO call will maximize log writer efficiency. Based on what I've said, you're probably already thinking about some of these options:

* Increase the log buffer size. This will reduce the likelihood of the log buffer filling forcing the log writer to immediately write possibly with less than its full 1MB potential.