c# - speed up through parallel.for -

need download lot of data (daily) oracle, delete data on sql server (in case of reruns) , paste data sql server sqlbulk copy. non parallel version :

     (int = 0; < curves.rows.count; i++)         {                 //download data oracle                 var data = getcurve(connectstring, impdate, curves.rows[i]);                 //delete old data in sql server                 deletesql(datetime.now, curves.rows[i]);                 //write sql server sqlbulk.copy                 writecurve(data, "dbo.t_curves"); } 

i wanted speed using parallel for, these steps dependent on each other. came (i have never used task factory before):

parallel.for(0, curves.rows.count, =>         {                 var taskload = task.factory.startnew(() => getcurve(connectstring, impdate, curves.rows[i]));                 var taskdelete = task.factory.startnew(() => deletesql(impdate, curves.rows[i]));                 taskdelete.wait();                 taskload.wait();                 var taskwrite = task.factory.startnew(() => writecurve(taskload.result, "dbo.t_curves"));   }); 

this halves time. halving time can expect? correct use of threading? can do? can include further speed ups?

update

just info, curves datatable definitions , m not writing or reading line line.

or should write this:

 parallel.for(0, curves.rows.count, =>         {                 var taskload = task.factory.startnew(() => getcurve(connectstring, impdate, curves.rows[i]));                 var taskdelete = task.factory.startnew(() => deletesql(impdate, curves.rows[i]));                 var res1 = await taskdelete;                 var res2 = await taskload;                 writecurve(res2, "dbo.t_curves");   }); 

q: correct use of threading?

a: yes, doing expensive work on 2 separate tasks, , , waiting them complete before carrying on. +1 @jugarr pointing out don't need task final bit of work

q: can do?

*probably no, sound of doing

• getting single row (from oracle)
• deleting single row (in case exists) in mssql
• adding single row mssql

some recommendations:

• batch up, don't work single row @ time.
• profile slowest part is, retrieving data, deleting it, or inserting? that:
  • working single rows, overhead of starting new connection massive. that
  • missing indexes when trying retrieve data
  • missing indexes when trying delete data because of referential integrity

what throughput rate getting? how many mb/s? each curve 1mb or 1gb?