Compiled by Eamonn Ryan based on the SAIRAC Johannesburg Centre Tech Talk on September 19 by Michael Young, mechanical engineer
Let’s visualise how the entire cooling system comes together in a modern data centre. This is Part 5 of a ten-part series.
Michael Young, mechanical engineer as part of a panel discussion at the Pan African Data Centre conference and exhibition earlier this year. © RACA Journal
At the core, we have racks housing our IT equipment, specifically servers equipped with direct-to-chip cooling technology. To effectively manage heat, we introduce a manifold—a long pipe with off-take pieces that distribute cooling fluid to the servers.
Manifold and cooling distribution unit (CDU)
Manifold functionality: The manifold serves as the central distribution point for the cooling fluid. Cold fluid enters the system, flows through the manifold, and is delivered to each server. After absorbing heat from the chips, the hot fluid returns to the CDU.
Cooling distribution unit (CDU): Within the CDU, a plate heat exchanger and pumps work together. Hot fluid from the servers enters one side of the heat exchanger, while cold fluid from the chiller enters the opposite side. Through this exchange, the hot fluid cools down, and the cold fluid absorbs that heat.
Understanding pod loads is crucial for managing data centre efficiency. IT equipment does not operate at a constant load, leading to variations in cooling needs.
- Variable demand example: On high-demand days, like Black Friday, servers experience peak loads due to increased online activity. Conversely, during off-peak times, demand drops, requiring less cooling. This variability necessitates a responsive cooling system. To adapt to changing cooling demands, we incorporate balancing valves within the manifold. These valves regulate the flow rate based on the cooling needs of individual racks.
- Balancing valves: Installed on both hot and cold water lines, these valves ensure that each rack receives the appropriate flow of cooling fluid, especially under maximum load conditions.
- Variable speed pumps: The CDU is equipped with pumps controlled by temperature and pressure sensors. These pumps can ramp up or down to adjust the flow rate based on real-time cooling requirements.
Feedback and control mechanisms
Effective monitoring and control are integral to maintaining balance in the system.
Temperature sensors: Supply water temperature sensors provide real-time data to determine if cooling capacity is adequate. If temperatures rise above a set point, the pumps increase flow to provide additional cooling. Conversely, if temperatures fall below the set point, the flow is reduced.
Three-way or two-way valves: These valves play a critical role in managing the flow between the cooling system and the chiller. If the system detects a drop in cooling demand, the valves adjust to reduce water flow, enabling the chiller to scale back its operation, saving energy and resources.
A balanced, responsive cooling system
In summary, the integration of advanced cooling systems in data centres, including the use of manifolds, CDUs, balancing valves, and responsive pumps, creates a balanced environment that can adapt to varying loads. This responsiveness is vital for maintaining optimal performance in a landscape increasingly dominated by AI and high-density computing demands.
By understanding these components and their interactions, we can better appreciate how to design and implement efficient cooling solutions that meet both current and future data centre requirements.
