DSXchange

Ok. Have you narrowed it down to the write stage?

Going to very high CPU is not always a bad thing - it means that you haven't spent a lot of wasted money on processing capacity. The sample code you've listed is going to chew up a lot CPU time, it can be optimized quit a bit. As I thought, the MOD() function is not the culprit. Are you doing the MO...

You choices are limited - reduce your CPU load or get more CPUs. What does vmstat show if you replace just your 4 MOD() functions with simple derivations and re-run your job? I would be surprised if just these 4 operations were consuming all that CPU time.

If you monitor your system (vmstat is common to all UNIX implementations, so use that) , does the %CPU used consistenly go over 90% with samples of 10 seconds when using the MOD() function? How many CPUs does your system have? I would think that just one transform stage but more levels of parallelis...

Craig - you are, of course, correct; there is no project-level switch allowing or disallowing Multi-instance jobs. I mixed that up with the project-level settings for RCP (runtime column propagation). The only place to actually enable multi-instance is in the job itself.

Replace your output load to an oracle stage with either a sequential file going to /dev/null or a copy stage with no output link. Re-run your job. How long does it take?
How many partitions does your database have and how many nodes are you running your PX job with?

What are your settings for Job Properties -> Performance for the row buffering settings? Also, the ipc_ problems could be symptomatic of the actual stage aborting and the 2nd process required for ipc no longer being there. You could try going to your project directory, changing into the &PH&...

Each job needs to be marked individually. You might be able to do this globally by exporting to .dsx or .xml and doing an editor global replace. Also, you should edit your sequences that call multi-instance jobs to ensure that they get the correct instance name. Checking the property in the sequence...

a) TX is a completely different product from DataStage PX or Server.

b) Premium membership is not necessary for DSXchange unless you want to see the full posts of certain posters and use the additional benefits that a premium membership will bring.

Why don't you create the .dsx file with binaries, then you wouldn't have to recompile.

No, the project level setting would only allow users to create multiple instance jobs. And not all jobs run correctly in multiple instances, i.e. if you have a hard-coded output sequential file.

So you are reading a table and transferring it straight to another table in the job? After the job has "hung" and you've restarted the server, did you recompile or use Director -> Job -> Reset/Cleanup Resources/Clear Status File? Does it always hang, does it hang if you write to /dev/null on a text ...

I am still unsure of what your actuall error or warning is. You need to make sure that the Oracle stage produces a format that the stage can use. Have you tried using an Oracle CAST... function to ensure that the format of the SQL SUM() function is decimal(38,10)?

Could you explain what didn't work - perhaps with an error message?

I think that there is a type of generation gap when it comes to using Windows as the OS for a heavy-duty commercial application; just as there remains a generation gap between the "heavy metal" mainframers and UNIX users. Just as UNIX was once considered to be a "toy" by mainstream data processing e...