Bill and the ‘Rat Pack’
 

Application cost optimization is one of TBM’s potential benefits. Once we understand the cost structure of an application (drivers, consumers, IT characteristics), we can consider how to optimize such costs. This is a question that business units often ask: “what can you do to reduce our costs?” Such initiatives—often called Application Rationalization, or ‘App Rat’ for short—show a higher level of maturity in TBM. In this episode, Bill the TBM Guy explains a method he recommends for such initiatives, which includes setting up a simulation for end-users to weigh hypotheses.

***

“I’m in trouble, Bill. I need your help.” Said Amy Rose, manager of a mission critical application family. “We met with the business, and Christina Davis, the BU representative, demanded that we reduce the costs of our most expensive application.”

“Are we talking about the availability processing engine here, Amy?” Asked Bill.

“Yes! According to your numbers, ‘APE’ is a multi-million expense every year. I understand little about the TBM engine. Can you help, Bill?”

“I think we can, Amy. We just put together a report we call Application Optimization.” Bill opened his TBM dashboard and clicked a button. The TBM engine opened a report, and sure enough, APE was very high on the list.

“What am I seeing there, Bill?”

“For every application instance, the report shows the monthly peak CPU utilization, and the minimum and maximum thresholds, i.e. how idle or busy servers in the farm can be. It also has the number of servers. We have added the application costs for reference.”

“Yes, having costs is something we find very useful in the TBM engine. What does that ‘Optimization Candidate’ flag mean, Bill?”

“We have defined two criteria to determine what applications are potential candidates for a cost optimization initiative. In the TBM jargon we call such projects ‘App Rat’, short for application rationalization.” Said Bill. “I am not sure I like that expression, that’s why I sometimes refer to it as the ‘Rat Pack’. That’s my particular sense of humor.” 

“And what are those criteria?” Asked Amy.

Bill said: “For now, we consider server farms with over 4 servers, and a monthly peak CPU utilization below the minimum threshold.”

Fig. 1: Capacity data for the APE application

“As you can see, the utilization was below the minimum threshold in that month. APE has hundreds of servers, and thus it qualifies as an optimization candidate. Do these numbers make sense to you, Amy?”

“Yes, this month was particularly low business-wise, so servers wouldn’t be running hot.” She said. “So, how much could I save if I got rid of some of those servers and returned them to the pool of available machines?”

 Bill said: “We are working on a prototype to help you do just that. Let me show you.”

 He switched to the development environment, and a different version of the report opened up.

 “Let's assume that we want to reduce by 10% the server footprint, that's 71 servers for APE.”

 “7% savings. See? That would be several hundred thousand dollars.” Said Bill.

 “Why not 10% savings? Wouldn’t it be logical?”

 “It would be wonderful if everything in life were that simple, wouldn’t it?” Asked Bill, with a big smile. “There are costs that do not depend on the number of servers, like database software. We allocate those based on the number of database servers, but there will be a few of those, even in a large application like APE. Also, application labor costs are relatively independent from the size of the farm. People report time against an application; only the IT infrastructure people’s costs would diminish, as these get spread across all servers.” 

“I am not sure I understand all of this…” Said Amy.

“Let’s look at it from another perspective.” Said Bill. “Do you have a target for the APE cost reduction?”

“They were talking about a 15% cost reduction. I have checked with our Capacity Planners and they reckon we could survive with 20% fewer servers.”

Bill showed Amy his screen. “What you’re seeing is a prototype I am building.  Let’s select APE.” He clicked on a drop-down field. “Let's enter the new number of servers. What should we enter, 600? That would be something like a 16% reduction.”

“Yes, please. Let’s see what it says.”

Bill entered 600 in the New Server Count box in the report and hit the Save button at the bottom;

Fig. 2 - Application Optimization Simulation report (1)

 “There you go, Amy. You would save about one month's worth of costs, or around 9%. Not bad at all, if you ask me…”

“How did you build the model, Bill? How real is it?”


“As real as it gets, Amy. I started taking out servers 10% at a time, remapping them to a dummy application. This prevents ripples in the cost structure that might have happened if I had just deleted them altogether. I let the engine calculate and noted the new cost of the application for each of the steps until I got to a 70% reduction. Beyond that, I thought we might get distortions. These are the results for what happened with APE.” Bill showed a graph plotting server reduction against cost savings:

Fig. 3 - App cost savings (y-axis) vs. Server reduction (x-axis) for APE

“Wow! It’s almost a straight line!”

“I repeated the process for the most expensive applications and we are now ready to share the prototype with you application owners. For the rest of applications, I came up with a similar algorithm and programmed it in the report.”

“I’ll go back to our Capacity Planners and discuss the plan with them. In previous talks, they suggested shutting down the servers, but keep them in store for a while, just in case.”

“Do that, Amy. We’ll try to work out a plan in case other application owners follow your steps. We shouldn’t have hundreds of servers idling, if you see what I mean…”

"Thanks, Bill. That was great!" 

***

“Hey, Bill. How did you build that simulation report?" Ellie Nakamura was another TBMA in Bill's organization. 

"After I built the first optimization report with fixed algorithms, I was looking for ways to allow users to perform what-if scenarios. I created an Editable Table out of the transform backing the original optimization report:

Fig. 4 - Generation of an Editable Table for the App Optimization Simulation report

I selected the columns I needed and added a new one that I called New Server Count. This is where end users will enter their data.

Fig. 5 - Additional column in the Editable Table

"What is the column for?"

"As I exercised with Amy, it allows the user to enter the new number of servers for the application. The report then calculates the new estimated cost using the algorithm I described."

"Could it be any number?"

"In principle, yes. The report doesn't check for any strange conditions. However, I know it accepts smaller or bigger numbers."

"Why is that important?"

"While not designed for that, we could project how much the costs would vary if we increase the number of servers. I just got an inquiry for such a project."

"And how did you go about that, Bill?"

"I walked the requestor through the report and had her add 46 servers to the farm. After hitting and clicking on the Save button, the report calculated the offset for the month, and the projections for the rest of the year."

"Do we promote that data at all?"

"No. We just leave it there. Data is just for the simulation, so it has no lasting value. If we needed, we could, but it's not required at this point."

"Thanks, Bill. Very clear!"

***

“Hey Bill. I hear you have some kind of application optimization engine…”

Bill hadn’t expected this call: it was rare that Jessica Harvey—the CIO—turned to the TBM Office for something.

“Yes, Jessica. We have been working with application owners to see how they can optimise their share of the IT infrastructure, server- and storage-wise.”

“Are you saying that we could save costs in IT with that mechanism?”

“Not directly, Jessica. As you know, server and storage devices are assets from a financial perspective. As such, they are subject to depreciation and amortization, D&A for short. We spread the costs of those assets over their lives, typically 5 years, so shutting down servers doesn’t automatically reduce the total IT costs.”

“But we would save electricity and cooling costs, at the very least.” Suggested Jessica.

“Maybe. I am not sure about the details of the contracts with the providers, but yes, in principle we could save some. Mind you, we’re talking about some hundred servers, out of several thousands.”

“So, what’s the point of the exercise, Bill?”

“Excellent question, Jessica. First, getting the application management people to be more conscious about the costs they generate. As you know, it’s all too easy drawing from IT resources as if they were free. Paraphrasing Lee Iacocca, ‘there’s nothing like a free lunch’. This is one step in that direction. Many times we just do things ‘because that’s how we’ve always done them’. That may lead to unnecessary redundancy and, hence, to excessive costs.”

“Yes, I get that. What can we do to save costs overall, then?”

“One thing we can do is to repurpose those servers. We can set up a process whereby those machines would go back to a pool, ready for the next application to use them.”

“But we’re trying to move to the Cloud, Bill. Wouldn’t that go against your suggestion?”

“Sure, Jessica. We should minimize purchasing additional hardware as much as possible. That’s one reason that gives additional merit to this idea of repurposing existing devices.”

“Anything else?” asked the CIO.

“There’s also the question of timing: we’re not migrating applications to the Cloud from one day to the next. The migration plan goes over three years, and the first pilot application migration will happen in Q2 next year. We’re going to keep needing on-premises infrastructure for a while.”

“So we’re not saving costs directly, but we may prevent shelling out cash in a near future…” Said Jessica.

“Exactly. Not only would we prevent additional purchases, but we also would decrease the backlog of hardware requests. As you know, that has been a sore point with the business units for quite some time.”

“Yes. It’s not magic, but it’ll sure help! Thanks, Bill. I’ll have a word with the infrastructure people and request that they sketch a plan to cover this repurposing exercise.”

“Thank you, Jessica.” Said Bill.

***
​​​

#cost-optimization

(c) Guillermo Cuadrado, 2022

All characters and events in this article are fictitious and any resemblance to organisations, locations or real persons, living or dead, is purely coincidental.