In Oracle 10g I believe that the value of LOG_BUFFER is determined dynamically, and can be as high as 14MB or thereabouts.
In Oracle 9i the size of LOG_BUFFER (as recommended by SAP) was around 1MB.
My understanding is that "log file sync" is time waited by shadow processes during the writing of the log buffer to online redo log files, which happens with every COMMIT, whenever the log buffer reaches 1/3 full, or every three seconds.
My question is why do we still expect average "log file sync" times in Oracle 10g to be below 15ms?
Say I have a system that has very high DML activity, but relatively low frequency of COMMITS. Then the Oracle 10g log buffer would be nearly 5MB before it was written to a log file.
Wouldn't this mean high average "log file sync" times would occur (higher than for 9i at least)? Or is this view of things unrealistic; do COMMITS actually happen too frequently to ever allow the log buffer to get so large?