Yes, at 8.0 LGWR always writes the entire "used" portion of the log buffer. At 8i, LGWR tries to avoid getting stuck waiting for incomplete copies into the log buffer, so it only writes up to the sync RBA (or for background writes, the RBA just before the first incomplete copy).

If you always commit intensively, then your log buffer size is an indifference because you will never fill 1/3 of the log buffer, and all writes will be synced. However, if you have intermittent pauses in that commit rate, then a modest log buffer size will help reduce the potential delay when those commits resume.

I have an application running against Oracle 8.0.6 in a E10K Box. The commit rate in the database is about 800 commits/seconds so I see a lot of waits for log file sync wait event (obvious). I would like to know what happens when LGWR writes to redo log files: is the entire log buffer written to redo log file when a session execute a commit statement? In the other cases LGWR has to write to disk (eg: 1/3 of the redo log buffer is full) is the entire log buffer written to disk or only a range of RBA's is written to disk?

In the case of LGWR have to write the entire log buffer after each commit I suppose I have set a small redo log buffer (256-512 Kbytes).