JDE B7334 on iSeries DB2 to be migrated on Oracle DB

Hi!

We have a data base of about 350 Gb on B7334 to be migrated from iSeries DB2 to Oracle DB in about 12/14 hours.

We are using the JDE standard tool. But the weakness of R98403 is that the API it uses, JDB_CopyTable via BSFN B9800200, seems to work on a record at a time.

We have already run in parallel UBE and for the big tables used data integrator tool (though not standard JDE). Is there anyone who has customized R98403/BSFN B9800200 to have an increase in the conversion speed.

Or any other suggestion? Thanks a lot for your input.

Paolo Daperno

[email protected]

I have customised a copy of the R98403 to remove the Windows record count screen. This allows it to run on an enterprise server which usually has processor capacity advantages over a fat client. Without the status screen the copy is a bit faster as well but the JDB_Copy API still generally copies record-by-record. If you are running your copies from 1 or more Fat Clients removing the status screen still helps as it makes it less risky that you will hit the enter key while one of the copy dialogs is active and cancel the copy. I also find that I can have more UBE's running locally without the system hanging. I have have gone as high as 30 concurrent from a single fat client.

Besides running multiple copies of R98403 to copy tables in parallel I do some tricks with datasources and views to carve up big tables like the F0911. I like my tables to be in chunks of no more than 2,000,000 records to fit within a 12 hour period. If I had a F0911 with 20,000,000 records I would want to run 10 pieces in parallel. To do that I would create 10 libraries on the iSeries (e.g. DCOPY01, DCOPY02, DCOPY03 ...) and associated datasources. In each library I would create a view that selects a year or year/period from the F0911 so that each library contains a view that returns about the same number of records and added together they represent all records. I then create 10 versions of the R98403 each pointing to one of the DCOPY datasources. I set the job to append records and not recreate the table. Finally I generate the F0911 empty to the target and then kick off all of the jobs in parallel.

I use R98403 exclusively for my platform migrations because it is supported. Some may disagree and want to go with native tools but I prefer the safety of knowing that if anything will be supported by Oracle for platform migrations it will be the standard JDE middleware tools. I have done a number of iSeries to SQL and iSeries to Oracle migrations using the above techniques with my largest at 750 GB taking 48 hours. 350 GB in 12-14 hours will be a tight fit. You can always create more copy versions and run them in parallel across multiple machines but you may still have network bandwidth constraints that may prevent that much data being migrated in that time.

I hope some of that helps.