SUN quad processors and Transformer 1

[stewm]

We are moving our production environment to a Sun Fire V440 Server with 4 1.28-GHz UltraSPARC IIIi processors. One of the options when creating data models is to sets whether you can use multiprocessors when they are available.

If your computer has multiprocessors, you can set this property to optimize performance during the read phase of column calculation and category generation

Default: False

Question:

How to you know for sure if Tranformer is using the all 4 of the processors. Or do they mean something different by multiprocessor use?

Thanks,


Mark Stewart
Senior Analyst
Consultants Club Corp.
Windsor, Ontario
Canada

http://www.consultantsclub.biz

 

[Draoued]

You can see in Windows Server the processor usage , and Transformer is for sure using as much as possible CPU.

One comment:
Current Transformer 7.3 mr1 is having a "bug", if you have more than 1 iqd where "enable Multi-processing" is ticked, the cube creation will stop with an error :-(

 

[stewm]

Thanks I can see the CPU usage fine in Windsows but there I mostly just have 1 processor on a PC. The SUN box has 4 unique CPU processors on one machine. I'm not sure but I thought the multi-processing option was for using some of each of the 4 processor.

Mark Stewart
Senior Analyst
Consultants Club Corp.
Windsor, Ontario
Canada

http://www.consultantsclub.biz

 

[Draoued]

Have a look to this one (you need a Cognos account)

 

[stewm]

FYI...

Got this from Charles at Cognos support...

1) How does the second process (transda) speed up the loading of the work file, what mechanism is it using?


The idea of the second process is that if a machine has 2 or more CPUs, some of the work involved in fetching the data from the source can be done in parallel with Transformer's processing. The second process will fetch the data from the source (possibly significant if an SQL database is being used), parse the records (significant for flat files) and calculate any Calculated Columns. With 2 or more CPUs, this can be done in parallel with Transformer because the second process will read ahead in the data
source.

2) If you turn multiprocessing off what effect does that have on the cube build?

If you turn it off, all the processing related to getting the data records is done within the same process. This may well slow down the cube build.

3) How does the trasnda process speed up the read, again what is the mechanism?


Reading can be speeded up because processing is done on a second CPU. This overlap will improve the total run time. But, it will be offset by the amount of work necessary to move the data from the transda process to Transformer. This is done through a pipe which is implemented (by the OS) as memory moves in shared memory.

4) What are the key differences between the 2 processors?


Transda contains the code within Transformer that reads data from a data source and massages it into data records that Transformer can then fetch fields from. This massaging includes scanning CSV files for the fields and calculating Calculated Columns. Transformer does all the modeling and Cube building.



Mark Stewart
Senior Analyst
Consultants Club Corp.
Windsor, Ontario
Canada

http://www.consultantsclub.biz