Java SP - ResultSet.updateRow causes 100x CPU consumption

Question

Oracle 11 Enterprise on Windows, with an out-of-the-box installation.
I select from a table. No where clause. No order by. If I use ResultSet.updateRow to change any column, the CPU consumption of oracle.exe jumps 100x. Here is the code.
Connection con = DriverManager.getConnection("jdbc:default:connection");
Statement stmt = con.createStatement(ResultSet.TYPE_FORWARD_ONLY, ResultSet.CONCUR_UPDATABLE);
ResultSet rs = stmt.executeQuery("select cola,colb from tab");
while (rs.next()) {
rs.updateString(1, "qwerd"); // comment out to save 99% of your CPU*
rs.updateRow();
rs.close();
stmt.close();
con.commit();
For my test case, against 10,000 rows, removing that one line changes the CPU consumption from roughly 7 CPU seconds to 0.07 CPU seconds. I do mean CPU time - not elapsed time. The CPU measurement come from v$sesstat.value. Task manager confirms oracle.exe using 100% of a cpu for the duration of the call.
I know that this example can be replaced with a single update statement. It is a gross simplification of a much larger proc, to replicate the problem.
Help?

Answer 1

Hi. For each row (when really updated), the DBMS is probably having to re-iterate
through the table in the same unqualified way your query does, to position itself
to the row you want to update. This whole thing is extremely inefficient, and I
understand that you've reduced the issue to the simple example, but this does
expose the critical fault of the algorithm. It breaks the number-one performance+
rule for DBMS applications+: Don't move the data. Don't bring whole swaths of
raw rows one-by-one to the client, just to find what you want to modify. Do the
work where the data is, in the DBMS, using stored procedures. Build your saw mills
where the trees are.
Joe Weinstein