AxcelPack vs PE

[datastage]

What advantages does AxcelPack bring in a multi-processor enviroment, and how does Parallel Extender go above and beyond those advantages?

Also, will AxcelPack continue to be available or is PE its replacement?

[bigpoppa]

DataStage,

I do not know what AxcelPath is so I can't answer your question. If I had more info on AxcelPath, I might be able to give you some info. Do you have a link for AxcelPath?

Thanks,
BP

[datastage]

If AxcelPack is installed, it adds an option in a jobs properties to 'allow multiple instances'. Then when you run a job in Director, a box appears for an 'Invocation Id', and it allows the user to run the same job multiple times at the same instant. For instance, I may have a job that processes four weeks of data and I pass it a parameter for which week to process. Without AxcelPack you need to either run the same job sequentially 4 times or create 4 copies of the job with different names and run them at the same time. In Director you will see the jobname with a period and the id you chose appended to it.

On a single processer box I'm assuming the performance would be the same as running the same jobs simulataneously but with different names and the AxcelPack advantage is just making the developers life much easier. I know that AxcelPack will have the multiple instances running as separate processes, but wasn't sure it would truly took advantage of multiople processors in the same way as Parrallel Extender would.

That's about all I know about AxcelPack, hopefully if I mistated anything or if there is more info someone else can clarify

[kcbland]

The Accel Pack is what Ascential called "job instances" with DataStage 5.1. You had to purchase this if you wanted to enable "job instances" capability. Since you could always make copies of a job and run multiple copies, Accel Pack was kind of a joke. In DataStage 5.2+, you didn't have to pay for the functionality.

All the functionality does is clean up some of the copying of jobs by allowing dynamic creating of copies at runtime. If you don't know what I'm talking about, go look at the Performance Analysis document I've posted on this site. It covers a discussion about "divide and conquer" techniques using multiple copies of a job.

If you're not on DataStage 5.1, Accel Pack doesn't mean anything to you. If you're on 5.2+, it's inherent functionality.

From the initial question, Parallel Extender is the ETL tool purchased by Ascential from Torrent. They integrated the GUI to create either jobs for traditional ("Server") DataStage or Parallel (Torrent's Orchestrate ETL tool). They are RADICALLY different job designs, frameworks, methods, and technologies.

-Ken

[bigpoppa]

From what has been posted, it seems like AxcelPath allows a user to do parallel dsrunjobs on a particular job.

This is **radically** different from what PX does. While running a DS Server job in parallel harnesses the power of partitioned parallelism, it does not harness the power of pipeline parallelism.

Remember, partitioned parallelism gives you the performance advantage of running discrete chunks of data through the same processes in parallel. Pipeline parallelism gives you the performance advantage of moving data to a downstream process as soon as the upstream process releases it.

Parallel Extender jobs should be faster than AxcelPath jobs because they access the Parallel Extender engine, which is tuned to run DS PX jobs in parallel at a very granular level - at the stage level. AxcelPath only provides parallelism at the job level.

[vasu]

[Vasu]

Any way the Ascential objective in PX is all the level of parallelism in one product.

[ray.wurlod]

Axcel PACK was the initial implementation of multiple instance server jobs. It was a separately charged part of the product at 5.1. It vanished (became free, with no need for separate licensing) at 5.2.

Parallel Extender is Ascential's name for the Orchestrate technology they obtained by acquiring Torrent Systems. Parallel Extender is an execution environment where processing (data partitioning particularly, but also pipeline parallelism) is automatically distributed over one or more "processing nodes" described in a configuration file.

Parallel Extender supports parallel jobs, which use an almost completely different set of stage types than server jobs.