Cognos8 and Transformer Performance

[brain1e55]

Hi.

Please, could you advice how can I tune cognos8 server to achieve the 100% utilization of all CPU's in a server. It is possible to have maximum load of all processors while some report _renders_ in HTML or Excel format? I tried to increase # of processes for Report service and Batch Report service in Server Administration interface, but there is no difference - only one CPU used, and the utilization peak freezed at the ~25%.

And one more thing. As I understand, Transformer's architecture doesn't support more than 1 CPU and there is no inverse proportion between number of CPU's and the cube generation time. Am I right? If so - it's a nightmare! :)

Thanks for your replies!

[mikegreen]

How many CPUs are in the C8 server?

Each C8 process will spawn a thread that will use only 1 CPU.  So if you have a 4 CPU system - one report will utilize only a single CPU.  However, if you have multiple jobs going or multiple users it will use all available (well, how ever many you have setup in the service settings).
Thus, NO it is not possible to have more than a single CPU execute a complex report. Yes, that sucks.

Transformer will support 2 CPUs.  On a 4 CPU box you should hit 50% utilization with Transformer building a cube.  There is a setting for this, but I can't recall it exactly.

Also, turn off HyperThreading in the BIOS.  HyperThreading makes Windows (I'm assuming you're on windows) think that there are twice as many CPUs in the server. This is bad for Cognos.

Regards,

Mike

[brain1e55]

Thanks a lot Mike. You cleared all the questions for me. And the things you've mentioned are really sucks!
What about other BI-platform products, i.e. Business Objects? Are they bad too at multiCPU processing?

[mikegreen]

I am not sure about bobj.

I would look at modifying the report or framework to make it render quicker... we changed somethings to make the database server do more work than the C8 box..

Regards,

Mike

[jguevin]

I saw this post and had to chime in, hope you don't mind...a couple of questions/comments:

1.  why is hyperthreading "bad" for Cognos?

2.  we are experiencing extreme slowness when it comes to cognos 8, especially analysis studio.  we understand that our cubes (now packages) are larger than the "best practice" according to cognos, but do either of you check out the task manager on the dispatcher where cognos 8 lives?  When we have 1 user using analysis studio the bibustkservermain uses 25% of the cpu.  This being said when we have 4 users hitting analysis studio slicing and dicing, our cpu is max'd out.  The machine itself is no slouch, dual p4 zeon 2.8ghz, memory 4gig.  Either of you experience this?

thanks
Jeff

[mikegreen]

1 - Hyperthreading makes the OS think that there are 2 processors for every single physical processor.  This is good for Windows and service applications that can be more efficient (and thus have better performance) by using multiple processors even though its still the same physical hardware..  Cognos applications are generally limited to a single processor (or at least what the OS says is a processor) per thread - like the BI processor in C8 or Transformer.  Transformer can actually use 2 processors.
The problem with hyperthreading is that Congos will peak at a single processors usage - so when the physical processor is split into two, Cognos is only using half of that processor.  It isnt smart enough to multithread over multiple processors for each job. 

2 - How large are your cubes?  What performance do you get with the Cognos sample cubes? Are the cubes you are building being built using the proper optimizer setting (meaning it is turned on.. if you installed Transformer from a C8 install download then it should be.. but check the transformer INI file and make sure)?  Series 7 cubes in C8 are really slow unless they are optimzed (reindexed probably is what occurs) for C8.
  Is only one CPU being used for multiple users? Or just for the single user? 

Regards,

Mike

[jguevin]

Hi Mike,

Our larger cubes run 116mg to 330mgs.  We are using transformer 7.3, and I have turned on the pcoptimizer.

As far as the cpu, it looks like 1 is used for multuiple users, but each user when running interactively consumes 25% of the cpu, so 4 users max at a time.  This would be 4 instances of the BIBusTKServer running.  We also notice multiple instances of java.exe running on the box which is consuming a lot of resources.

Breaking it down even further, when running the AS reports in interactive mode, when we change rows or columns or measures on views of our largest cubes,  (you see the hour glass working, then drawing)  On the 'working' part it is hammering the dispatcher, then on the 'drawing' part it is hammering our local desktop, sometimes max'ing 100% of the desktops cpu.

I have heard from other users, as well as cognos people, that there are a lot of performance concerns with Analysis studio in Cognos8.

Thanks
Jeff

[mikegreen]

Jeff,

  Ouch.. I think that might be typical... What MR are you running of C8? We got a little performance increase by
going to MR1 from the original (sloppy) C8 release... I am not working with it day to day anymore - is MR2 out?

  I've heard the same on the performance issues...many people are disappointed..

Cheers,

Mike

[jguevin]

Mike,

we are running on MR2.  Hopefully next major release is a bit faster.