Evolving Trends and Technologies Shaping Data Center Strategies
by Andrew Lui
In recent times, we've observed intriguing trends impacting the development of data center strategies. Building on my earlier piece about data center transformations adapting to the needs of the modern digital consumer, it's now fitting to explore how advanced technologies are catalyzing change in data center design and construction.
The Importance of a Clear Data Center Strategy
Data centers, utilized for an array of applications, often house mission-critical workloads. Establishing a data center demands an optimal location with robust utility infrastructures, sufficient expansion space, and equipment tailored to its specific purpose.
Without a clear data center strategy, the business may lack the capacity to evolve with their customer’s requirements for new services or applications and may cause heavy losses in the future. Even the world’s largest data center owners are not immune to this. In December 2022, Meta halted a $ 343 million data center construction in Odense, Denmark, which was due to a change in data center strategy to focus more on Artificial Intelligence. Not only that, it was found out later that Meta was reworking the design of 11 under-development data centers and optimizing them to better support AI workloads. This adjustment coincided with the rise of ChatGPT and a surge of new AI applications in the consumer mainstream.
Recent Developments and Technologies Driving Data Center Transformation
Advances in artificial intelligence and machine learning are stretching the limits of existing data center capabilities. Implementing innovative architectures and technologies allow data centers to fulfill emerging requirements for capacity, latency, automation, and sustainability.
Data centers for the new age of applications are expecting to build much larger compute clusters, interconnecting tens of thousands of servers. Latency becomes a very critical issue as services move closer to real-time responses, as seen in the case of ChatGPT and other chatbot search engines. The growing size and complexity of data centers will also pose challenges for manual management, necessitating the implementation and enhancement of automation.
Building sustainable data centers have become a very important topic of late due to the number of resources required to maintain them. Governments are imposing more regulations on data centers to track and review data center energy performance. Moreover, numerous major data center operators have committed to increasing their use of renewable energy and optimizing their data center infrastructure to improve power usage effectiveness.
Optimizing Data Center Infrastructure
The data center infrastructure is composed of servers, storage equipment, network equipment, power systems, and cooling systems. Optimizing a data center involves improving the efficiency and cost-effectiveness of these components.
Servers are the computing hardware within the data center. When we think of a data center “crunching data”, that is being done on servers. Virtualization is increasingly prevalent in data centers, allowing multiple users or computers to share a single physical server. This improves resource utilization and reduces costs.
Efforts are underway to enhance architectures for distributed servers, where compute resources can be distributed across multiple data center sites. This approach can be taken further by separating and pooling server components such as CPUs, DRAMs, I/O, NIC, and disks, enabling better resource utilization.
Data center storage consists of equipment and software used for storing data. There are three types of storage used in data centers. Direct-attached storage (DAS) stores the most frequently used data to be directly accessed by a device and is commonly located close to its CPU. Network-attached storage (NAS) provides data storage to multiple servers and is accessible as long as it’s connected to the same network.
High-performance computing and distributed architectures in data centers drive advancements in storage technologies, such as Compute Express Link (CXL). CXL is an open industry standard interconnect between host processors and devices. It is designed to support heterogeneous processing and memory systems with applications such as artificial intelligence, machine learning, analytics, and high-performance computing.
Networking equipment, such as switches and routers, facilitates data transmission within the data center, connecting servers, storage, and end users. Networking speeds and capacities are growing alongside data center workloads. As servers continue to adopt 25G and 50G interfaces, it is fueling the growth in 400G switches as well. It is projected that 400G switches will account for 11% of the total data center switch market share by 2025, according to a leading industry analyst Omdia.
Networking is also going through a renaissance with the introduction of a distributed disaggregated chassis architecture such as that developed by UfiSpace, which uses interconnected fabric and line white boxes to build virtualized switching chassis with up to 691Tbps switching capacity. The fabric white boxes utilize cell switching technology to minimize latency and increase the number of connections.
The power infrastructure consists of multiple sources of power, power distribution units (PDUs), and uninterruptable power supplies (UPSs). All of this with the single goal of keeping the data center operating continually without interruption. Data centers consume substantial electricity and can put a strain on power grids, impacting nearby communities.
Sustainability efforts are a priority for data center operators who want to improve their power usage effectiveness (PUE). PUE is a metric used to measure the overall energy efficiency of data centers with a ratio of the total amount of power consumed to the power delivered to IT equipment. A PUE ratio of 1.0 means all of the energy consumed is used to power its IT equipment and not wasted in other areas. That being said, according to Uptime Institute, the 2022 global average data center PUE was 1.55, which was a slight improvement over the 2021 average of 1.57.
Data centers are adopting renewable energy sources such as solar, wind, geothermal steam, and hydroelectricity from rivers to reduce PUE. Additionally, the trend of using power-efficient servers and switches continues, along with the development of new and innovative cooling methods.
Cooling infrastructure prevents equipment from overheating, which is crucial to prevent downtime. Efficient cooling strategies are a key focus in data centers, as they will also significantly impact PUE. As servers, switches, and storage become more powerful, they consume more energy and generate more heat. Therefore, the more efficient the cooling system, the lower the PUE.
Several cooling methods are being explored including new rack designs to maximize natural air cooling. There are also new concepts being proposed for liquid cooling and immersion cooling, with immersion cooling being more popular. The Open Compute Project has laid out guidelines for immersion cooling and has been conducting much proof of concept with data centers and equipment vendors.
Collaborative Approach to Data Center Strategy
Adopting a clear data center strategy not only empowers data centers to meet evolving customer demands but also paves the way for future infrastructure optimization. The strategic approach to data center planning holds significant implications for businesses and industries, necessitating the evaluation of current infrastructures, a comprehensive understanding of business requirements, and the selection of a suitable transformation approach.
Given the pivotal role of technology, businesses are encouraged to actively collaborate with technology vendors and providers to develop a robust data center strategy that aligns with their unique goals and objectives. This collaborative approach will enable businesses to adapt to emerging trends, leverage new technologies, and ensure the efficiency, reliability, and sustainability of their data center operations.
For UfiSpace networking solutions that can improve your data center architecture, contact our sales team.