All of the O’Reilly books ... the ones with animals on the cover ... are done that way.

Nope.

Thanks all. Here is the SQL which generates the XML file. However it takes days to generate the XML file for 20 million records. I do have proper indexes on the tables. Could you please provide your ideas to generate this XML. Which one is the fast method to build this XML?. 1. Extracting the data to CSV file takes about 15 minutes using ETL and generate the XML using Perl. 2. Otherwise writing a SQL by joining all the normalized tables and generate the XML file by calling the SQL within the perl.

select xmlserialize(document
       XMLElement("transactiondetails", 
                 XMLElement("rltp_id", rltp_id), 
                 XMLElement("rltp_name", rltp_name), 
                    (SELECT XMLAGG(XMLElement("product", 
                                  XMLElement("prod_id", p.prod_id), 
                                  XMLElement("prod_name", p.prod_name)
+, 
                        (SELECT XMLAGG(XMLElement("customer", 
                                      XMLElement("cust_id", cust_id), 
                                      XMLElement("cust_name", cust_nam
+e), 
                                 (SELECT XMLAGG(XMLElement("account", 
                                            XMLElement("acc_id", acc_i
+d), 
                                            XMLElement("acc_name", acc
+_name), 
                                            XMLElement("acc_balance", 
+acc_balance),
                                            nvl(
                                              (
                                               select xmlconcat(
                                                        xmlelement("ac
+c_type", at.acc_type),
                                                        xmlagg(xmlelem
+ent("acc_code", at.acc_code))
                                                      )
                                               from acctype at
                                               where at.rltp_id = a.rl
+tp_id
                                               and at.prod_id = a.prod
+_id
                                               and at.cust_id = a.cust
+_id
                                               and at.acc_id = a.acc_i
+d
                                               group by at.acc_type
                                               )
                                             , xmlconcat(xmlelement("a
+cc_type"),xmlelement("acc_code"))   
                                            ) ,
                                        (SELECT XMLAGG(XMLElement("tra
+nsaction", 
                                               XMLElement("txntrack", 
                                               XMLElement("txn_id", t.
+txn_id), 
                                               XMLElement("txn_amt", t
+.txn_amt),
                                               (
                                                select nvl(
                                                         xmlagg(xmlele
+ment("txn_code",tt.txn_code))
                                                       , xmlelement("t
+xn_code")
                                                       )
                                                from txntype tt
                                                where tt.rltp_id = t.r
+ltp_id
                                                and tt.prod_id = t.pro
+d_id
                                                and tt.cust_id = t.cus
+t_id
                                                and tt.acc_id = t.acc_
+id
                                                and tt.txn_id = t.txn_
+id
                                               )
                                               ))
                                               ORDER BY t.txn_id
                                               ) 
                                        FROM transaction t 
                                        WHERE t.rltp_id  =a.rltp_id 
                                        AND t.prod_id   =a.prod_id 
                                        AND t.cust_id   =a.cust_id 
                                        AND t.acc_id=a.acc_id))
                                        ORDER BY a.acc_id
                                        ) 
                            FROM account a 
                            WHERE c.rltp_id=a.rltp_id 
                            AND c.prod_id=a.prod_id 
                            AND c.cust_id=a.cust_id))
                            ORDER BY c.cust_id
                            ) 
                        FROM cust c 
                        WHERE p.rltp_id = c.rltp_id 
                        AND p.prod_id = c.prod_id))
                        ORDER BY p.prod_id
                        ) 
                     FROM product p 
                     WHERE p.rltp_id = r.rltp_id )        
                    ) indent) AS xml 
FROM 
  rltp_mngr r 
  WHERE rltp_id='10';
[download]

Assuming that your dump to CSV is ordered like you showed in the original post, I'd probably go with dump to CSV and then write some perl to read the CSV line by line and build up the XML using XML::LibXML. Perl is pretty quick at that sort of stuff. I have some code that's all perl calling MySQL and for stuff that's a few thousand records it generates a bunch of HTML files (not too different from building up XML) plus separate text audit files by the time I've lifted my finger off the return key. For stuff that's reading a few thousand lines of CSV and doing inserts into a database of millions of lines (including some lookups to decide how to cross reference things) it takes a few minutes.