Problem with partitioning

[ramesh_c]

Hi All,

I am running a small jobwith sequential file as input ,Transformer &Oracle enterprise stage as output.

I applied partitions bykeeping entire partition on Transformer stage and round robbin on oracle enterprise stage . But upto transformer the data flow is very fast and showing about 4500 rows/sec. But at oracle stage it is slowing up by showing 85 rows/sec when i viewed it through job monitor.Why it is slowed i cant understand. Is the partition on oracle stage is wrong?.
Another Question is What is the partition that is generally applied on Stored procedure stage,

Thanks,
Ramesh.

[balajisr]

I applied partitions bykeeping entire partition on Transformer stage

Why did you use entire partitioning?

[bucks]

Please change the partition to hash at transformer stage and auto at oracle stage.

[ArndW]

After changing your partitioning methods you might not get much greater speed, but at least your data will be correct. What options are you using in the Oracle stage - particularly the load method.

[ramesh_c]

Hi Arnd W,
I am using the Round Robbin option in the oracle stage .

Thanks,
Ramesh.

[ArndW]

Round robin will work at the output stage, but the entire partitioning you specified earlier in the data stream has corrupted your data (except for cases where you have only 1 node in your APT_CONFIG file). Correct that error and don't change partitioning schemes unless you need to. Why not do a round-robin earlier on in the job and leave everything downstream of that set to "Auto" (which is the default).

[ray.wurlod]

Are you using an Oracle Enterprise stage? If so, are you using direct write or Upsert? If Upsert, what upsert mode have you specified? Are there any (many) warnings in the job log?

Stored Procedure stage often runs in sequential mode, which means that you're more interested in a collection algorithm than a partitioning algorithm. If you need to keep data sorted use Sort Merge, otherwise use Auto.