For years we have watched enterprise pursue digital transformation and looking to save money, leading to an increase in considering leasing facility for IT infrastructure instead of owned and operated by themselves. What Kind of data center would be right for enterprise IT infrastructure? In this post, we will examine the key comparison metrics for selecting a data center.
Comparing data centers
There are various comparison factors such as scale, performance, efficiency, cost, geographical location, and natural environment, but in Part 1 of Comparing Data Centers, we will explain the key comparison metrics in terms of scale and performance. The scale and performance of data centers can be compared primarily through metrics such as Capacity, Redundancy, and Efficiency.
Capacity: This metric represents the maximum capacity of IT infrastructure that a data center can accommodate. Traditionally, the scale of a data center has been mainly indicated by area or the number of racks. Since almost all data centers supply rack power of 2~3 kW, the measurement of scale has expanded to include IT load (the electrical capacity available for IT infrastructure) along with area and rack quantity due to the diversity of rack power in recent times.
Redundancy: This refers to the availability at which the IT infrastructure housed in a data center are operated without downtime. Data centers provide a certain level of availability to their clients. Availability is mainly assessed through two indicators. The first is the Tier rating, which classifies the backup level of infrastructure facilities (mechanical and electrical facility with multiple distribution path). The second is uptime. the guaranteed annual availability of a data center represented as a percentage. The uptime is ensured through a contract known as the Service Level Agreement (SLA).
Efficiency: Efficiency refers to the energy efficiency of data center infrastructure facilities (mechanical, electrical, communication, fire protection, etc.). With the proliferation of commercial AI, the demand for GPU servers has increased. AI workloads consume more than 10 to 20 times the power of regular workloads. As nearly 50% of the cost is invested in mechanical and electrical facilities compared to conventional buildings, the operational efficiency of these facilities directly affects cost savings.
Metrics for Capacity, Redundancy, and Efficiency
The capacity, redundancy, and efficiency of data centers can be quantitatively assessed through several indicators. Capacity indicators include the area of the data center's computer room, the number of racks or rack power, and IT load. Availability is primarily represented by the Tier of the entire center's availability level, and N indicators are used for the availability of individual equipment within the center. Efficiency is often indicated by PUE (Power Usage Effectiveness), which assesses how efficiently a data center uses electricity. PUE is calculated by dividing the total load (total electrical capacity the center can use) by the IT load (electrical capacity available for IT equipment), with lower values indicating higher efficiency.
Capacity
The capacity of a data center refers to its scale. It's a measure of how many racks it can accommodate. In the past, when rack power was standardized at 2kW, it was possible to compare the scale of data centers between centers based on their area. However, with the diversification of rack power, including power consumption exceeding 50kW due to machine learning and AI, the assessment of data center scale is increasingly based on IT load rather than physical area. The fact that recent large data centers are handling IT loads of over 20MW reflects this trend.
Data Center Area: Measured in pyeong or sqm (square meter). A larger area indicates a data center with favorable scalability.
Number of Racks: The quantity of racks that can be accommodated is proportional to the area. A higher quantity of racks is indicative of better scalability.
Rack Power: Measured in kW per rack. The upward trend in rack power continues to be a significant evaluation metric for data centers.
IT Load: Measured in kW or MW. The total amount of power that a data center can supply to IT equipment.
[fig.1. average rack density growth trend] 
Source : Uptime institute 의 rack density is rising (2020)
Redundancy
Among the performance indicators of a data center, availability is more important than anything else. This is because it indicates the level at which the IT equipment in the computer room can operate without interruption. In order to ensure that IT equipment operates optimally, the data center operates numerous facilities such as machinery, electrical systems, fire protection, and security. In the process of operating these facilities, situations may arise where operations are disrupted. The three main cases include unexpected failures, regular testing and inspections, and planned maintenance for the replacement of aging equipment or components. A data center that can achieve uninterrupted operation of IT equipment in all three situations is evaluated as providing a high level of availability.
Tier : A measure indicating the level and grade of uninterrupted operation of IT equipment, often representing the overall availability level of the center. A higher tier indicates a data center designed to provide higher availability.
Please let me know if there's anything else you need or if you have any more text you'd like me to translate!
|
Rating |
TIER I |
TIRE II |
TIER III |
TIER IV |
|
Basic Capacity |
Redundant Capacity Components |
Concurrently Maintainable |
Fault Tolerance |
|
|
Detail |
Single path for power and cooling. Must be shut down completely to perform maintenance. |
Single path for power and cooling + backup components Maintenance to the power path and other infrastructure require a system shut down. |
Multiple paths for power and cooling, along with backup redundancy. Able to maintain without interrupting operations. |
Zero points of failure. Able to maintain in the event of downtime. |
|
Components |
N |
N+1 |
N+1 |
2N |
| Path |
N |
N+1 |
N+1 |
2N |
N: N represents the availability level of key facilities that make up the center. N refers to the basic capacity and distribution paths. It is expressed as N+1, 2N, etc., based on the N value. More information about the N approach will be covered in upcoming electrical series posts.
Efficiency
PUE (Power Usage Effectiveness): An indicator used to assess how efficiently a data center is designed and operated in terms of electricity usage. It is calculated by dividing the total load (the total electrical capacity the center can use) by the IT load (the electrical capacity that IT equipment can use). In simple terms, it reveals what portion of the total electrical capacity of the center is supplied to IT equipment. Typically, it ranges between 1 and 2. A value closer to 1 indicates a center that is designed and operated for higher efficiency.
This content is a copyrighted work protected by copyright law and is copyrighted by DAOU Technology.
Secondary processing and commercial use of content is prohibited without prior consent.
Featured images via gettyimages.
References
1. Uptime Institute : Rack Density is Rising (2020)
