Re: How large should a redo log file be?

Big enough to support you peak processing times/activity without causing a performance problem and small enough not to cause an undo delay in bringing the database back up after a hard crash.

The answer is it depends on your environment. I suggest you look at the log creation times to make sure that during peak on-line and batch times that the time between log switches is not too close. During our MRP run our logs were filling up in only one minute. Every time Oracle switches logs it checkpoints which means all dirty (changed) buffers get written to disk and the header blocks get updated with the checkpoint number (SCN). This impacts the performance of all running processes.

There is an init.ora parameter to force checkpoints by time instead of by blocks (which most shops set larger than the physical log sizes). Using a time of one hour would force Oracle to checkpoint during slow activity periods so you would be limiting the amount of recovery data while you size your redo to support peak processing.

I was once told by an Oracle trouble call handler that an average of one log per hour, ie, 24 per day is a good 'rule of thumb'. But the real answer is what size will work best for your shop.

Mark D. Powell -- The only advise that counts is the advise you follow

                             so follow your own advise.