DSXchange

Write cache can not be used if the hashed file is larger than the amount of allocated cache (or that percentage of a shared cache specified by the DCMAXPCT and other percentage parameters).

You can tune the amount of cache, but not to more than 999MB.

Use an Aggregator stage. Group your column and count your column.
Refer to Chapter 17 of Parallel Job Developer's Guide (parjdev.pdf) for information on how to use this stage type. It works best if the input data are sorted.

Hi We are also faced same kind of problem. What I did directly used UNIX and then killed the process using process id. From director you can able to see the job process id. Really if you got any idea please share with us. My idea is: Don't kill processes at UNIX level. There is a child, parent, gra...

Even so, it's the same answer as the one Beeram gave. If you think about it, the only efficient way that grouping can be achieved is by sorting, then detecting change. The difference between "don't sort (previously sorted)" and "don't sort (previously grouped)" is, therefore, primarily about documen...

Can't answer that until we know whether or not you've enabled read cache. I have heard it argued, primarily for file dictionaries, that modulo 1 forces the contents of the hashed file all to be loaded into memory at first access (because the whole group needs to be scanned on first access), effectiv...

Whatever you do to guarantee unique IDs will involve single-threading, and therefore a "choke point", so you have to make it as quick as possible. You could modify the SDK routine and make it more efficient. You could rely on a sequence or a SERIAL data type in the target database. If you're using i...

You can do whatever you like with variables in the controlling job. When you attach the controlled job and set its parameter values using DSSetParam, these values remain constant for the entire run of the controlled job. You must pass the parameter values to the controlled job. If you're working for...

On the server machnie take a look at the code generated by compiling the Transformer stage. This can be found at RT_BP3164/JOB.2000320632.DT.1340850175.TRANS1 - what's happing in and around line 57?

When you use MINIMUM.MODULUS or static hashed file, the hit of allocating disk space is taken when the hashed file is created, rather than while it is being populated. The best way to speed population of a hashed file is to enable write cache, which means that the "writes to random groups" occurs in...

The Job Sequence is a GUI from which job control code is generated.
Go to the Job Control tab in job properties and examine the code. Compare it to how you would have coded it. This gives good insight into best strategies for using Job Sequences for those who used to write job control code by hand.

This "solution" will only work for 30 days. Then you will have to buy a licence for the ODBC driver, or for some other third party UNIX-based ODBC driver that you elect to purchase. The branded ODBC drivers that ship with DataStage are licensed only for use with DataStage stages, not for use with BC...

Best practice dictates that every record loaded is marked with some form of identification of the load run, perhaps a batch ID, perhaps a timestamp. If you do this, unwinding inserts becomes a simple constrained DELETE statement. If you've stage a before-image of the table you're loading, unwinding ...

Maybe not stop it, but litigate all the profit away (and then some).

Only lawyers get rich. And savvy bookmakers.

Learn patience, grasshopper!

The underlying database is based on UniVerse, which is now owned by IBM. UniVerse manuals can be downloaded for no charge from IBM website.