A checkpoint is now done every MAX_LOG_SIZE / 2 instead of every MAX_LOG_SIZE, so that the transaction log doesn't grow as large.