Apptio for All

Apptio for All

A place for Apptio product users to learn, connect, share and grow together.

 View Only
Back to Library

Challenge 6: Improve physical server cost model [+++] 

Mon December 03, 2018 10:26 AM

Challenges are designed to help sharpen your TBM and Apptio skills. See title for rating: [+] Easy   [++] Moderate   [+++] Challenging

Its the FINAL TBM Pursuit of 2018. Challenges one, two, three, four, and five are in the books.

Here's your last chance to earn a piece of the pie. So grab your thinking caps, because Chris is ending with a bang. 

A correct answer is worth 30pts and a 2018 TBM Pursuit game piece 

Submit your answer by Dec 31.

@Chris Davidson says...

Recently I reviewed the portion of my ATUM-compliant Cost model which estimates fully burdened physical server costs.

Here's an excerpt of the relevant cost model objects:

 

 

At the IT Resource Towers (ITRT) object level, I'm using TBM Taxonomy v2.1 (details here).

My IT Resource Towers object is backed by a data table containing 41 rows which correspond to each tower and sub-tower combination listed in the taxonomy.

 

As expected, my Data Center tower cost allocates from ITRT to Data Centers object.

Then it allocates from Data Centers to Physical Server object, weighted by # CPU cores per server.

(So for instance, a server with 8 cores receives twice as much Data Centers cost as a server with 4 cores.)

 

Also as expected, my Compute tower cost allocates from ITRT to Physical Server object, weighted by # CPU cores per server.

In my screenshot above, I have separate allocation lines for Unix and Windows, but I could combine these if I wanted to (by setting up an Operating System direct data reference between the two objects, to ensure cost does not get mixed between OS's).

 

But three issues weigh heavily (pun intended) on my mind:

 

1. Server depreciation cost allocates from Fixed Asset Ledger to ITRT to Physical Server, but since I weight solely by # CPU cores (with no allocation filters), some of this depreciation cost is probably being allocated to servers which are already fully depreciated, unfairly driving up their estimated cost.

 

2. Server depreciation cost (again, originating from Fixed Asset Ledger object) winds up allocating across multiple servers as weighted by # CPU cores, but the number of cores seems somewhat unrelated to the amount of depreciation each server should receive. I have many 8-core servers whose initial purchase price was lower than some of my 4-core servers, for example.

 

3. My data center power bill correctly rolls up through the model to ITRT to Data Centers to Physical Server object, and I understand that the majority of a server's power is used for its CPU(s). But different CPUs use different amounts of power, and besides, my server CPUs aren't 100% active all month long. Weighting data center power cost by # CPU cores per server therefore doesn't seem fully defensible.

 

What improvements can I make to my Cost model

to address all three of the issues above?

 

 





#ApptioforAll

Statistics
0 Favorited
2 Views
0 Files
0 Shares
0 Downloads

Comments

Apptio Community Member
Thu January 10, 2019 11:05 AM

Yes Paula Foster not exactly the same but we're definitely doing a learn while playing series. @Meghan Johel and I are meeting in a few weeks to nail it down :-) 


#ApptioforAll

Apptio Community Member
Tue January 08, 2019 12:27 PM

I think these challenges are really great, it surprises me on how much i have learnt and continue to learn, hope this continues in 2019 or is there thoughts on a similar game where we can continue to learn and gain points!!   


#ApptioforAll

Apptio Community Member
Thu January 03, 2019 11:34 AM

Well done @Michelle McGuire


#ApptioforAll

Kyle Castro
Wed January 02, 2019 03:59 PM

Hmmm this actually makes me ponder about our current approach to allocating server costs... CPU Cores are what we're currently using at the moment, since we have no other numeric fields to go by.

 

Thanks for this, @Chris Davidson. This is like premier success advice stated in the form of a challenge. Kudos!


#ApptioforAll

Apptio Community Member
Wed January 02, 2019 11:33 AM

Congratulations to our winners for round 6!  You have all now been awarded the 6th Community Pursuit badge!

 

Thanks, @Chris Davidson for hosting our 2018 Community Pursuit challenges!


#ApptioforAll

Jenny Franklin
Wed January 02, 2019 10:46 AM

Happy New Year and hoping to see more challenges - these are so much fun and so beneficial!!!


#ApptioforAll

Chris Davidson
Wed January 02, 2019 10:16 AM

Congratulations to 9 of you who submitted a correct answer:

Mark Salib
Michelle McGuire
Tony Wong
Karen Lifsey
Steven Young
Paula Foster
Axel Burkert
Jenny Franklin
Jason Edward Tucker

 

Several of you rightly questioned the merits of "improving" the cost model:

1. Are these changes worth the trouble? (See: [++] TBMA Challenge: Assess impact of wrong allocation strategy )

2. Will reconfiguring the cost model potentially hinder our ability to report on specific kinds of results? (See: Challenge 3: Optimal cost model configuration [++]  )

3. Do we even want the cost model to more accurately reflect server cost, or are we instead hoping to use TBM methods to shape user demand for server-reliant apps and services? (For instance: maybe we want to impose extra cost on lesser-used, aging hardware to encourage users to migrate to newer platforms.)

 

For now let's assume:

1. Yes.

2. Not noticeably.

3. Accurate cost estimates are our highest priority. (Maybe we'll use a separate Charge model and a service rate card for cost showback, instead of using our actual Cost model results.)

 

First, we need some way to identify which servers have already fully depreciated.

If we have this information in our fixed asset ledger, we could carry this information up from Fixed Asset Ledger to IT Resource Towers (ITRT) object.

For example, adding a sub-sub-tower or metadata column to one or more new ITRT rows (such as Compute - Servers - Depreciation) lets us keep depreciation separate from other kinds of server cost (labor, data center, projects).

 

Alternatively, we could create an entirely separate Depreciation model (which may or may not resemble our existing Cost model), removing depreciation entirely from the Cost model. This grants us greater modeling flexibility but would require us to adjust either our metric formulas or our related reports (which are only expecting Cost model). 

 

Then we can solve two of our original issues: We can add new depreciation-specific allocation(s) inbound to Physical Server object which weight by initial purchase price (or by estimated monthly depreciation, if our server data contains this or if we can retrieve it from another data table) instead of weighting by # CPU cores.

 

If we don't think # CPU cores is the best way to weight data center power costs (and again assuming cost accuracy is more important than demand shaping), we could instead consider % CPU utilization (if available) as a weighting factor.

 

We could also look up per-server or per-CPU power characteristics (using manufacturer data sheets) and combine this with % CPU utilization data to create estimates of actual power consumed (in watts), then use it as the cost weighting factor (as power-specific subset of cost moves from Data Centers to Physical Server object).

I have not yet seen any cost models which go this far in the name of modeling accuracy, but it's a solid example of "What would I do if I had unlimited data and time?"

 

 


#ApptioforAll