Divide and Conquer. Understanding the Impact of Modular Design Approach on PUE.

I. Introduction

When designing a physical Data Center, the phrase “Divide and Conquer”, has never been more truth than ever, in order to future operate a Data Center facility with a high level of Energy Efficiency. The concept of a Modular Design approach, basically means aggregating RACKs in PODs (Point of Deliveries) with their associated Site Infrastructure components. It can have a major impact on PUE, particularly when the IT Load total utilization is lower than 50%.

However, their benefits might go beyond the Energy Efficiency/Sustainability umbrella by impacting:

a. Improving Data Center overall Reliability and therefore increasing its Availability

b. Decreasing the Need for Redundancy, or i.e. dedicating its function to most critical PODs

c. Improving TCO by implementing PODs on demand

The downside of this approach is that might impact the maintainability function, and the need for better monitoring and operational readiness, because of the additional infrastructure.

By dividing the Data Room into multiple delivery units of Site Infrastructure components, will tackle the inherently inefficient Data Center growth, especially in the early stages of the IT Load growth process.

Nevertheless, is also important to understand that regardless of a Highly Energy Efficient Data Center Design approach, PUE is going to be impacted positively or negatively by Data Center operations. Implementing measures like a good Air Management program, operating closer to the ITE Intake Air Temp upper limits, controlling IT Load demand and Cooling System response adaptation, among many others measures, could have a huge additional impact on PUE over its best design values.

II. Impact of IT Load utilization on PUE

In most traditional Data Centers, the typical Data Center Design PUE curve, as in Figure 1 showing the “Chart Anatomy of a Design PUE Curve of a Typical Non-Modular Data Center”, demonstrates that the Data Center Growth process is inherently inefficient. The Figure 1 basically states three different stages, according to the correlated PUE values with respect to the IT Load utilization:

• Inefficiency Range: 0% - 50% IT Load. Worse PUE values, according to its Design PUE Value.

• Efficiency Limbo >50% - 70% IT Load. Undefined Efficiency.

• Efficiency Range > 70% IT Load. Best PUE values, according to its Design PUE Value.

The best PUE value is obtained at 100% IT Load utilization, and it can be referred as the Design PUE value. The Data Center Energy Efficiency comparative loss, in terms of PUE and according to its Design PUE value, is significantly higher during the first half of the total IT Load utilization growth of the Data Room. Typically, after reaching 50% of IT Load utilization, the comparative loss according to its Design PUE value, starts decreasing considerably. The best relative efficiency levels are gotten after reaching beyond 70% of the total IT Load utilization of the Data Room

Figure 1.

Chart Anatomy of a Design PUE Curve of a Typical Non-Modular Data Center

By reaching beyond 70% of the total IT Load utilization, means that the correlated PUE value can be considered in the Efficiency Range, according to its Design PUE Value (PUE at 100% IT Load). However, it does not necessarily mean that the resulted PUE value in the Efficiency Range would have a global generally accepted performance considered as Good or Efficient.

The range in between the Inefficiency Range and the Efficiency Range, might be referenced as the Efficiency Limbo. It would typically start from 50% up to 70% of IT Load Utilization, but these range limits can vary from one Data Center (Energy Pattern) to another. Nonetheless, many typical NonModular Data Centers might adjust pretty well to that efficiency pattern shown in the Figure 1.

There are many different design considerations, in order to built a very solid energy efficient facility. But, for the purpose of getting an Energy Efficiently Operation with a mostly continuous good PUE value, the way the Data Room is architected will have a major impact during the IT Load growth process. By implementing a Modular Design Data Room, the Site Infrastructure components can be activated and deactivated, according to the IT Load demand, in an efficient manner. Producing a remarkably lesser comparative energy loss with respect to a Non-Modular Design. Especially, during the early stages of the IT Load Growth, and all the way up to 50% IT Load utilization, referenced as the Inefficiency Range.

III. Modular Data Room Design Vs Non-Modular Design demonstrations

In order to demonstrate the impact of a Modular Data Center design approach on PUE, three different Case Scenarios were modeled for Energy Efficiency comparison purposes:

1. Non-Modular Data Room (One Site Infrastructure Elect/Cooling Component per Room)

2. Modular Data Room w/four PODs (One Site Infrastructure Elect/Cooling Component per POD)

3. Modular Data Room w/eight PODs (One Site Infrastructure Elect/Cooling Component per POD) Not Shown.

Assumptions:

1. Total Data Room Capacity (IT Load) for Each Case Scenario: 1000kW

2. Case Scenario w/four POD: 250kW IT Load Cap. per POD

3. Case Scenario w/eight POD: 125kW IT Load Cap. per POD (Not Shown)

4. PUE Design Value for Each Case Scenario: 1.615

5. Site Infrastructure Efficiency Patterns: For learning purposes and simplification, Energy simulations and calculations for each Site Infrastructure Component, were done considering an identical Efficiency Pattern according to its type of equipment, regardless of its size. That’s why the PUE Design Value is the same of each case scenario.

Measuring the Energy Efficiency Comparative Loss with the Design PUE Curve

In order to measure how efficient a Modular Data Room Design with respect to a Non-Modular Data Room Design approach can potentially be, the Design PUE Curve of each case scenario were projected for comparison purposes. Figure 2, shows the Design PUE Curve for the Non-Modular Data Room case scenario, and Figure 3 shows the Design PUE Curve for the Modular Data Room with four POD case scenario.

Figure 2.

Design PUE Curve of Non-Modular Data Center – Case Scenario

Figure 3.

Design PUE Curve of Modular Data Center (4 POD) – Case Scenario

The first important difference between the two Design PUE curves, is that the Modular Data Room with four POD (Figure 3), produces a non-continuous PUE curve. After a POD is completed in full, it may have two different PUE values at the same IT Load. The higher PUE value, might be referenced as the Jump PUE value, and is the result of a new activated POD with its associated Site Infrastructure (UPS, Cooling, etc), before the Implementation of new IT Equipment (same IT Load). After the completion of IT Load in the initial POD A, the worst PUE values of any POD in a Modular Design Data Room, are the Jump PUE Values. Nonetheless, these Jump PUE Values are significantly lesser than the correlated PUE values of a Non-Modular Data Room at the same IT Load utilization.

In the Figure 4, both Design PUE Curves for Non-modular and Modular Data Room case scenarios, are shown to showcase the PUE gap among them. During the first half of the total IT Load utilization growth, is when the greatest impact can potentially be produced on PUE, by the Modular Data Center (4 POD). For instance, by having a Modular Data Center operating with an IT Load utilization below 50% for a long period of time, would produce a lot of energy savings that might ranges from 29% to 59% depending of the time spent at certain IT Loads.

Figure 4.

Design PUE Curve Comparison between Non-Modular DC and Modular DC

Basically, every PUE value at different IT Loads in the Modular Data Center (4 POD) case scenario, is mostly significantly better than in the Non-Modular Design approach scenario, with the exception (very slight difference) of the Jump PUE value when POD D is activated at 75% IT Load utilization. The differential PUE values between both approaches, at early-stage IT Load utilization of 10%, 20%, and 25% are 3.59, 1.79, and 1.38 respectively, potentially saving up to 48% energy in an evenly distributed growth case scenario, during that ramp-up time.

The Figure 5, shows the corresponding Data Center total energy consumption in a 30-day convention (monthly) aggregation, at ten percent IT Load Utilization increase per month, for both Non-Modular Design and Modular Design Data Room case scenarios. The greatest potential energy savings can be obtained during the first half of the IT Load utilization growth, ranging from 59.82% savings at 10% IT Load, to 21.93% savings at 50% IT Load. After passing 50% IT Load, the energy saving rates fall importantly to 9.68% at 60% IT Load, and to 8.90% at 70% IT Load. At 80% IT Load
and beyond that point, there would be no potential savings since PUE values are almost identical all the way up to 100% IT Load utilization.

Figure 5.

Energy Consumptions and Potential Energy Savings of Modular DC with respect to Non-Modular DC

The same scenario shown in Figure 5, would potentially produce around 15% energy savings, during that total 10-months ramp-up time period. By having a Modular Data Center with 8 PODs, the additional savings would be only 2% in that same period, over the Modular Data Center with 4 PODs.

Using POD Partial PUE (pPUE) Design Curve for comparison purposes

Partial PUE can be a great tool for comparison purposes, particularly to analyze the level of potential efficiency of a POD and other Modular approaches, with real or projecting case scenarios. The Design POD pPUE curve, is the resulting graph considering only the associated POD’s Site Infrastructure Components (UPS, RPP, CRAC/AH), simulating their energy consumption as a function of the POD IT Load.

Figure 6.

Design Partial PUE (pPUE) Curve of POD A, B, C, D.

The Figure 6 shows the Design pPUE for every POD (A,B,C,D) with the same IT Load Capacity (250kW), which produces the same pPUE curve pattern. The POD Design pPUE value of 1.529 would be reach at 100% POD IT Load utilization. As the POD pPUE is calculated considering only the Site Infrastructure Components serving the POD IT Load, the corresponding POD pPUE curve and Design pPUE value, must be better than a hypothetical POD PUE curve and Design PUE Value, as if it were the only Data Room of a Non-Modular Data Center, since it would include other infrastructure energy consumptions outside the boundary of the POD Infrastructure.

The Figure 7 shows the progression of IT Load utilization on each POD, in order to visualize the Design pPUE curve with the corresponding PUE values at each POD of the Modular Data Center case scenario. The PUE values (Modular Data Center) are the same projected in the Figure 3 and Figure 4. Starting with the initial POD A, the IT Load progresses sequentially to the next POD and so on (POD A, POD B, POD C, POD D).

The POD A, as the initial POD to be activated, is the only one which the PUE values are higher than the POD pPUE values. The Differential values, are the infrastructure energy consumption outside the POD’s boundary.

Figure 7.

Comparison of Design PUE curve of PODs (A.B,C,D) vs. Design pPUE of POD

After the IT Load utilization progresses over the next PODs (B,C,D), most PUE values are better than the POD pPUE values, until the 80% POD IT Load Utilization is reached. The PUE values at 90% and 100% POD IT Load are slightly worse than the correlated POD pPUE values for every POD (A,B,C,D). After POD A has reached 100% POD IT Load, the remaining PUE values are below PUE= 2.0; with the exception of the activation of POD B at 0% POD IT Load utilization (PUE=2.17), and at 10% POD B IT Load utilization (PUE=2.01).

That comparative analysis, also demonstrates that the shared energy infrastructure for PUE calculation across PODs, gets significantly diluted after passing the initial POD A. Being only contrasted at 90% and 100% POD IT Load utilization on each POD (B,C,D).

IV. Conclusions

Operating and Controlling the Data Center Energy Efficiency is not a simple endeavor. It requires substantial amount of knowledge (multi-discipline) and continuous operational work. However, by designing a Data Center thinking in the sustainable operations, a Modular Data Room Design fits with an Energy Efficiently operation, that may even impact other areas. Among other benefits, a Modular Data Center will:

• Improve considerably the Energy Efficiency during the first 50% IT Load utilization, and extending the savings at least at the 70%+ IT Load utilization.

• Improve the Reliability thanks to the distributed approach.

• Potentially help to reduce or eliminate Redundant components. Depending of POD’s IT Load criticality.

• Improve IT Load higher densities optimization, by implementing POD with dedicated technologies (i.e. Liquid Cooling, etc.)

• Improve the TCO thanks to the Site Infrastructure Components On-demand implementation.

• Improve overall DC operations and readiness.

As Data Center Energy Efficiency and Sustainability technologies evolves over time, Divide and Conquer might be the timeless motto to be used, in order to keep energy efficiency operations under control.

Author: Jorge A. Gil - jagil@seresdc.com