Want to hire me? For availability and scheduling please email info@arcepm.com

29 July 2014

Calculation Manager, BSO Planning, and ASO Planning combine for an awesome ASO Essbase procedural calculation hack -- Part 2

Introduction

This is part two of a three part series on Calculation Manager, ASO Planning, its relationship with BSO Planning, and Essbase ASO procedural calculations.  Part one covered all sorts of interesting things you can do with, oddly enough, BSO + Calculation Manager.  There is a reason for introducing that first, and then this post on making ASO procedural calculations fast.  I’d encourage you to stick around for part three which ties them both together in a most unusual way.  With that, let’s get into the details of making ASO procedural calcs fast, fast, fast.

Joe Watkins’ genius

How often do you get called genius?  If you’re anything like yr. obt. svt. it isn’t all that regular of an occurrence unless that term is within the context of, “You have a genius for screwing things up” or “You’re a genius at prevarication” or “I’ve met geniuses and you are no genius”.  

Joe Watkins (with whom I have only ever “met” via email and Network54) is a genius, at least when it comes to ASO Essbase, because the approach Joe came up with is…genius.

Why do I use the word genius?  Simply because:
  1. His approach is 100% undocumented.
  2. It is not an intuitive solution at first glance, but on examination it is not only obvious, it’s !@#$ing awesome.
  3. It solves a problem you could drive an HGV double-articulated lorry through.
  4. It is fast, fast, fast, fast.
  5. It is a total hack.

And oh yes, it is part two of my three part series on Calculation Manager, BSO Planning, ASO Planning, and ASO Essbase procedural calculations.

While this blog post stands on its own for Essbase-only practitioners for the technique alone,  I will explain why you will at least want to combine it with the CDF information I gave in part one even if the words “Hyperion Planning” never cross your lips.  Hyperion Planning geeks will have to read all three parts to get all of this hack (and yes, I contributed something to this, so it isn’t all a case of steal from others to get a good solution).

The problem(s) with ASO Essbase procedural calcs

It’s really very simple and very devastating at the same time – ASO procedural calculations do not ignore empty level zero member intersections but instead consider all of them.  Where’s there constituent data, the Essbase values the result; where there’s none, Essbase leaves the result blank.  For us BSO geeks, this is 100% not the way BSO Essbase works by default; in BSO, no blocks = no result unless we force Essbase to do so.  If only there were a way to make ASO Essbase behave the same way…

What this ASO Essbase behavior means is that procedural calculations, unless they are very tightly constrained in scope, can be agonizingly slow.  And even that tight targeting of the calculations can be a roadblock – do you always know where in a database a calculation should occur?  Maybe you could write a series of small scope calcs in a reporting application, but that would be very difficult to do if there is an element of budgeting to the application.

And in fact, I’ve understated the problem – even in a pretty small ASO database a procedural calculation can take a very, very, very long time.  Proof?  Read on.

The database

I stumbled (as is my usual wont) into this as part of a presentation.  I was trying to write a currency conversion calc to mimic, sort of, what happens in BSO Planning as part of an ASO Planning Kscope14 presentation I gave with Tim German.  I should also mention that Dan Pressman was a big help in the building of the dimensions.

The dimensions


By ASO standards it isn’t much.  Of course, by the lights of BSO, it’s huge – more on that in a moment.

The logic

Oddly, ASO Planning does not create all of the dimensions required for a multiple currency database.  They expect you to do so.  No problem, I thought, I’ll simply create similar dimensions to what exists in BSO Planning and go from there.  

Dimensions

Here’s how the fx relevant dimensions look when compared to a standard Planning-derived BSO fx plan type:
BSO
ASO
HSP_Rates
Fx Rates
Account (HSP_RateType and children)
Account (Fx Rate Type and children)
Currency
Currency
Product (Entity)
Product (Entity)
N/A
Analytic

For you Planning geeks, the Product dimension is the Entity dimension, Currency was automatically created by Planning (although corresponding fx conversion logic was not), and Analytic, and Fx Rates are custom dimensions I created to support ASO fx.
Account
Accounts stored the difference between Average and Ending rates.  This is just like BSO Planning.
FX Rates
I don’t have a HSP_Rates dimension but this is the same thing, mostly.
Currency
This dimension came directly from Planning itself.
Product (Entity)
The concept of tagging the members with UDAs is the same as BSO Planning.

BSO code

Want the code?  Create a multicurrency Planning app and then have Planning generate the calc script.  I’m just showing the screen shot to give you a feel for the logic.

All that the code is doing is:
  1. FIXing on level zero members
  2. Testing for UDAs assigned to the Entity dimension
  3. Multiplying the Local input value by the fx rate / by the USD rate (which is always 1)

As this is BSO, after the fx conversion, an aggregation is needed but not shown.  Of course ASO won’t require that aggregation code as it does that aggregation on the fly.

ASO first attempt

I wrote this in Calculation Manager (remember, I was trying to do this for Planning) but the logic is exactly the same in MaxL.
Execute calculation

Note the SourceRegion – in this case it’s all individual members because I was trying to calculate just one rate.  I would at least have to open up the fx member set if I were to calculate more than one.
fxtest.csc
This is a one line currency conversion formula.  
This whole approach is ugly and not easily exapandable, but it serves to illustrate an almost literal translation from BSO to ASO logic.
Success, if you can call it that.  
After entering the rates (sent from a Smart View ad-hoc sheet into the Essbase database as ASO Planning doesn’t push rates the way BSO Planning does), I entered one data value to be converted from Sterling to Dollars as per the above code.  That is one as in a whole number, greater than zero, but less than two.  How long do you think it took to run?  A second?  Half a second?  Something else?

How about 6,000 seconds?  Let me repeat that in words, not numbers: six thousand seconds or sixty hundred seconds or one and two thirds hours.  To convert one data value.  See the problem with ASO procedural calculations?

How long did the same database (just about) with the same amount of data take to run in BSO?  The total elapsed calculation time was 0.025 seconds.  So much for the might power of ASO compared to poor old obsolete BSO.

The fix

The key to BSO’s speed is that BSO does not consider all of the possible level zero member intersections, it only considers the sparse member combinations that have data.  In ASO terms, it only calculates based on the non-empty member intersections.  There are NONEMPTYTUPLE and NONEMPTYMEMBER commands in MDX but unfortunately they are not part of the execute calculation and execute allocation (the two and only two ways to run ASO procedural calculations) grammar.

NB – Oracle say this is coming to Essbase but when is tbd.  That will be (some day) great for those of us who are on the latest release, not so much for everyone on 11.2.3.500 and before.

So how can we get NONEMPTY functionality in ASO if it’s not part of the commands?  Enter, finally, Joe Watkins’ technique again.

The problem with NONEMPTYTUPLE

NONEMPTYTUPLE (I used that instead of NONEMPTYMEMBER) can only be used in an outline member formula.  Member formulas are (quite naturally) in the outline.  The member formula fires at all levels, and in the case of an fx rate calculation, is only actually valid at level zero.

This is a bit of an impasse – we know the problem with procedural calcs is the NONEMPTY issue, fx rate calculations only make sense at level zero, MDX has a keyword to address this, but only in member formulas and member formulas fire at all levels, not just level zero.  What to do?

Back to the Genius of Joe Watkins

Instead of trying to fight Essbase, Joe came up with a really clever way of using existing ASO functionality.  I read about his approach to fx over on Network54 in the beginning of 2013 and, since I was doing BSO Planning (that’s all there was) at the time, filed it away for future reference.  I also thought he was nuts for saying things like, “This is the future of ASO.. BSO will be dead in 5 years.. (my prediction)....”  Now I’m not so sure.

What he did

Joe:
  1. Created a member formula that contained his fx logic
  2. Stuck a NONEMPTYTUPLE keyword at the top of the formula
  3. Ran an ASO procedural allocation (not calculation) that 100% copied his member formula to a stored member thus harnessing ASO’s fast non empty performance but keeping the data calculated only at level zero
  4. Enjoyed success, the good life, and that warm glow inside that only comes from helping others

I may be slightly exaggerating number four, but one through three are the truth.

NB – The example here is a fx calculation, but this approach works for any and all level zero calculations.

Here’s how I did it

Additional member

In the Analytic dimension, I created a calculate-only member called MTD USA.  It contains the member formula to calculate fx conversion.

MTD USA’s member formula

Note the NONEMPTYTUPLE command that makes the member formula only address non empty data.

The CASE statement is a MDX version of the BSO currency conversion calc script.

Execute allocation

It’s all pretty simple from here on, thanks to Joe.  All I need to do is kick off an execute allocation in MaxL, set up my pov aka my FIX statement, identify the source (Local) and target (USD).  By not defining a spread range other than USD, Essbase copies everything from MTD USA in Local to MTD in USD.

Did you see the $5, $6, and $7 in the code?  If it’s MaxL, it can be driven through parameter variables.  Think about how you might use that in Planning given the last post’s review of @CalcMgrExecuteEncryptMaxLFile.

How fast is it?

On my VM with a limited set of data (I have finally ordered that 1 TB SSD but have yet to install it so I am constrained for space) I observed the following calculation times:
Process
BSO
ASO
X Fast
Allocate
106
3
35
Fx
400
1.2
333
Aggregate
1,772
N/A
N/A
Total
2,278
4.2
542

The allocate and the aggregate times are interesting, but the biggest overall difference is in fx – it’s over 300 times as fast as the equivalent BSO outline and data.  Look at that, ASO is faster than BSO, if it only considers non empty data.  

And now, thanks to Joe’s technique, it can.  A hack, most assuredly, but a glorious one.  I have an upcoming ASO budgeting application that I was dreading because I couldn’t figure out how to quickly do its rate calculations (no fx involved).  Now I know how to do it, and quickly.

This technique is in a word, awesome.  Yeah, I take some stick for overusing that word, but 300 times the speed of BSO is sort of remarkable.  All of us who use ASO procedural calcs owe Joe a huge round of thanks.

So what’s left?

Part three of this series will bring together:
  1. Running MaxL from BSO while in an ASO Planning application with Calculation Manager
  2. The fast ASO procedural calcs you just read about
  3. How to use this in Planning (and even Essbase)

I know it’s all been a bit long but there’s a lot of information to impart and it took me freaking forever to figure out how to tie it all together – there’s no reason to see why explaining it should be any faster.  

:)

Be seeing you.

8 comments:

Sathish Radhakrishnan said...

Hi Cameron,

I tried exactly the same stuffs as a part of POC (using NONEMPTYTUPLE on the member formula) and calling it through allocation process.

I submitted the POC to my client about a month ago. Unfortunately, I dropped this procedure due to a performance degradation, since it was executed at all the accounts level. Is there any specific parameters or set of accounts you are using to execute this procedure or is it getting executed at the entire DB level? Please clarify

Thanks,
Sathish Radhakrishnan

Sathish Radhakrishnan said...

Hi Cameron,

I tried exactly the same stuffs as a part of POC (using NONEMPTYTUPLE on the member formula) and calling it through allocation process.

I submitted the POC to my client about a month ago. Unfortunately, I dropped this procedure due to a performance degradation, since it was executed at all the accounts level. Is there any specific parameters or set of accounts you are using to execute this procedure or is it getting executed at the entire DB level? Please clarify

Thanks,
Sathish Radhakrishnan

Cameron Lackpour said...


Satish,

I had the same problem when I expanded the range of Accounts.

At least in the case of my test database, it seemed to be a function of how many Accounts (so dynamic compression dimension) were accessed.

What I did was take the logic and simply repeat it about 7 times (yes, really) always going after a different tree in Accounts.

Believe it or not, that was really quite fast although profoundly ugly.

Sometimes ugly wins. :)

Regards,

Cameron Lackpour

Sathish Radhakrishnan said...

Thanks for the excellent post and response.

Unfortunately, I'll not be able to do so with few accounts.

It was much faster when I used partitioning to transfer the data from the BSO (level0) for the entire set of accounts. The entire data set went within 1 min and tied out.

Regards,
Sathish Radhakrishnan

Sree said...

I have created a bug with the Essbase folks and see how long will they take to resolve this.

Sree Menon

amarnath said...

Thanks for the excellent post Cameron.
This is what exactly I did around 2 years back for one of the implementation where we did a BSO to ASO conversion and we came up with all possible options...Reading your post brought me back those memories...we had sleepless nights working on this conversion :)

But the only difference that we had was we didn't use the procedural calcs but rather used it purely with dynamic scenarios

Leanne said...

Just curious. Is there security so the users cannot see the accounts calculated at each level with the MDX formula that gets copied to the stored member? We do similar logic in BSO Workforce so we can utilize dynamic calcs but want to only run on level 0 without the weight of relying on calc scripts. We have to hide these "driver" accounts to eliminate confusion.

LuigiO said...

Has someone been able to get around the level 0 restriction in PBCS? Is there a way to make it work in PBCS?