Page 1 of 1

View Construct Issue PMR 86469,756,000

Posted: Tue Feb 28, 2012 8:58 pm
by jim wood
Guys,

Do you know if there is any difference (apart from unicode) between view construct in 9.1 and 9.5.2? How does it handle the out of page view selections?

Thanks in advance,

Jim.

Re: View Construct

Posted: Tue Feb 28, 2012 9:06 pm
by ellissj3
Jim,

I am unaware of this on a technical level, but we've seen problems using this function in 9.5.2. Using this function caused a crash of our service.

Of course, I tried using this on our largest, most rule-driven cube... :) I don't know what happened under the hood here, but I am interested to hear the responses to this.

Re: View Construct

Posted: Tue Feb 28, 2012 9:08 pm
by jim wood
I have a bit if a weird one were I have a cube of around 400MB and another of around 14GB. When using view construct for both cubes in both cases memory increases by 2GB???? Previously when they ran it it didn't behave this way. Memory went up but not to the extent.

Re: View Construct

Posted: Tue Feb 28, 2012 9:15 pm
by ellissj3
I saw similar issue where we had an operating RAM at 50Gb (as a service). This ended up eclipsing 195Gb, causing a service crash.

Re: View Construct

Posted: Wed Feb 29, 2012 3:48 am
by lotsaram
jim wood wrote:Guys,

Do you know if there is any difference (apart from unicode) between view construct in 9.1 and 9.5.2? How does it handle the out of page view selections?

Thanks in advance,

Jim.
Jim what exactly do you mean? ViewConstruct caches values in from the view in memory, nothing more. I don't think unicode is relevant (other than additional memory use in general), and what is "out of page view selection"?

Re: View Construct

Posted: Wed Feb 29, 2012 4:14 am
by nick_leeson
Highly Speculative but worth a shot !!

I am not sure but have a look at this document http://www-01.ibm.com/support/docview.w ... wg21460595

"Internal testing comparing like tests comparing TM1 9.5.1 to TM1 9.5.2 with PI have shown the TM1 9.5.2 Server to allocate between 10 – 30% more RAM than the TM1 9.5.1 server. Some of this increased memory use is due to the internal structures created by PI, however the additional throughput of transactions – and their impact on View Cache – is also a contributing factor. IBM recommends planning for 30% more memory usage with TM1 9.5.2 than with prior versions of TM1."

It would be nice to know whether if PI is enabled in all the instances which are having this issue.

Re: View Construct

Posted: Wed Feb 29, 2012 2:51 pm
by jim wood
lotsaram wrote:
jim wood wrote:Guys,

Do you know if there is any difference (apart from unicode) between view construct in 9.1 and 9.5.2? How does it handle the out of page view selections?

Thanks in advance,

Jim.
Jim what exactly do you mean? ViewConstruct caches values in from the view in memory, nothing more. I don't think unicode is relevant (other than additional memory use in general), and what is "out of page view selection"?
I know what it does so thanks for that. I also know what the impact of unicode is (which is why I put apart) so thanks also for that. When I say out of page view selections I mean the ones at the top that are not physically within the view page area were you have to have one element selected? Does that help? I can upload a screen dump if you like??

Re: View Construct

Posted: Wed Feb 29, 2012 2:53 pm
by jim wood
nick_leeson wrote:Highly Speculative but worth a shot !!

I am not sure but have a look at this document http://www-01.ibm.com/support/docview.w ... wg21460595

"Internal testing comparing like tests comparing TM1 9.5.1 to TM1 9.5.2 with PI have shown the TM1 9.5.2 Server to allocate between 10 – 30% more RAM than the TM1 9.5.1 server. Some of this increased memory use is due to the internal structures created by PI, however the additional throughput of transactions – and their impact on View Cache – is also a contributing factor. IBM recommends planning for 30% more memory usage with TM1 9.5.2 than with prior versions of TM1."

It would be nice to know whether if PI is enabled in all the instances which are having this issue.
It doesn't have PI enabled as it is a BI model and there isn't a great deal of writing going on. I good punt though and thanks.

Re: View Construct

Posted: Wed Feb 29, 2012 4:05 pm
by lotsaram
jim wood wrote:.... When I say out of page view selections I mean the ones at the top that are not physically within the view page area were you have to have one element selected? Does that help? I can upload a screen dump if you like??
Only the title elements selected by index are included in the cached stargate view. It doesn't matter if a subset is assigned with other elements in it, if you want to cache for more than one filter element in one or more of the filter dimensions then you would have to cycle through by index and do more ViewConstructs.

If you are talking about the feature in TM1 Web and BI where you can have a filter/title dimension that isn't visible to the user then I am pretty certain that it would be treated the same as any other title element. I don't see how this could be any different since the view construct isn't something that happens on the UI level, it is just a background job to pre-calculate an array of values that belong to the view.

Re: View Construct

Posted: Wed Feb 29, 2012 6:24 pm
by jim wood
lotsaram wrote:Only the title elements selected by index are included in the cached stargate view. It doesn't matter if a subset is assigned with other elements in it, if you want to cache for more than one filter element in one or more of the filter dimensions then you would have to cycle through by index and do more ViewConstructs.
That pretty much covers it cheers. That's always what I understood to be the case but I was unsure whether it had changed down the version. Thanks.

Re: View Construct Issue PMR 86469,756,000

Posted: Mon Mar 05, 2012 3:21 pm
by jim wood
I have moved this to the bugs section as IBM have been able to replicate the issue,

Jim.

Re: View Construct Issue PMR 86469,756,000

Posted: Tue Mar 06, 2012 9:43 am
by moby91
What version of the TM1 9.5.2 software are you running ?

Are you already running TM1 9.5.2 FP1 HF5 or newer ?


TM1 9.5.2 FP1 HF5 includes two patches addressing defects of the TI fct ViewConstruct logged as APAR PM49801 and PM54252:

(1)
https://www-304.ibm.com/support/entdocv ... wg1PM49801
PM49801: TM1 VIEWCONSTRUCT IS NOT WORKING - OUTOFMEMORY EXCEPTION <<< MEMORY_TEMP_POOL_EXCEEDED >>>

Error description
After unloading a cube, any attempt to reconstruct the view fails.

(2)
https://www-304.ibm.com/support/entdocv ... wg1PM54252
PM54252: EXCESSIVE START UP TIME...UNKNOWN

Error description
Following the Initial Calc Chore - comprised of 'CubeProcessFeeders' and 'ViewConstructs' - the duration of the chore has jumped from 30 mins to 1 hr to over 1 day.

Re: View Construct Issue PMR 86469,756,000

Posted: Wed Mar 07, 2012 3:25 pm
by jim wood
Many thanks Moby. They are running the latest 9.5.2 hot fix pick install. They have not installed any hot fixes beyond the pack. As for the fixes / errors thanks for making me (and the community) aware of them but the issue we have doesn't directly correspond to either but it is close. You never know either fix may address the issue indirectly but just to clarify....

1) We are not unloading the cube before running view construct.
2) We are not running any other TI command other than view construct, so process feeders isn't applied.

I will however recommend to the customer that they apply the latest hot fix just in case,

Jim.

Re: View Construct Issue PMR 86469,756,000

Posted: Wed Mar 14, 2012 3:27 pm
by jim wood
Just a quick note to let you know that the development team have indentified this as an issue with the software. Neither of the fixes raised by Moby helped. A fix for this is due in the next FP which is slated for release in April. If however they manage to fix the issue before hand a hot fix will be released. Either way I will add the details to this post,

Jim.

Re: View Construct Issue PMR 86469,756,000

Posted: Mon Apr 16, 2012 1:42 pm
by jim wood
Guys,

According to IBM this issue has been fixed with 9.5.2 FP2. For a reason that is beyond me that didn't add this PMR to the fix list,

Jim.