sybperl-l Archive

From: AGHZHONG at aol dot com
Subject: How can I get the speed I need?
Date: Jul 20 1999 12:17PM

I need to move massive amount of data (7 million rows) from one database to 
another.
The two databases have different tables and different columns.  I cannot do a 
straightforward bcp in.

The only approved method right now is to bcp out from one database and read 
the data file row by row and then insert into various tables to the other 
database.  
This has to be done as fast as possible.

For a bulk estimate on how long the process will take,  I tried a simple 
case.  A 2000
row data file is read row by row and then converted column by column and 
finally insert into a (one for now) table using SybTools (built on top of 
Sybase::CTlib) ExecSql function call one row a time.  The result is very 
disappointing. I can only insert about 20 rows a second.  (Most of time is 
indeed spent on insertion,  not on those string manipulations at all.)  

Then I tried to group a batch (2000 rows) of records together and wrap the 
batch with "begin transaction" and  "commit transaction" using the same 
function call.  The result seems to be slightly better.  I can now insert 
about 30 rows a second.  This is still too slow for what I need to do.

Next thing I want to try is to build bcp-able array in memory as I read in the
input file and bcp the array into the table in batch using the BCP tools that 
came with Sybperl. I don't know this can be any faster either, as BCP 
functions are 
built on top of DBlib and probably not much faster than the CTlib calls.

Any suggestions on how I can speed up? If I write a C program that targets to 
what
I need to do.  How much faster can I go?  Anyone has any experience on that?

Thank you
Heather