Abstract
In recent years, with the rapid development of big data and cloud computing technology, large Internet companies have taken the ocean with natural heat dissipation advantages as the location of data centers. However, there are still many complex factors to consider in the construction of data centers in seawater. This paper is based on the background of the first submarine data module in China, exploring the optimization of server density by balancing physical space constraints with thermal limits.
Introduction
Traditional land-based data centers are resource-heavy, consuming vast amounts of land, electricity, and fresh water for cooling. Submerging data centers in the ocean offers a "natural" cooling solution, but it introduces a complex engineering trade-off: density vs. temperature. If too many servers are packed into a sealed underwater container, the heat generated will exceed the rate at which the seawater can whisk it away, leading to hardware failure. The goal of this research is to find the "sweet spot"—the maximum number of servers a single container can house without overheating, considering both geometric constraints and the physics of natural convection.
Methodology
The study determines the optimal capacity $N$ by applying a "minimum of" programming model: $N = \min(n_1, n_2)$, where $n_1$ represents physical packing limits and $n_2$ represents thermal limits.
Physical Packing ($n_1$)
To find the maximum number of servers that physically fit, CAD software was used to simulate three different placement orientations within a cylindrical container. Because rectangular servers cannot perfectly fill a circular cross-section, unavoidable gaps are created. The orientation of the servers drastically changes the capacity.
Thermal Modeling ($n_2$)
To find the heat dissipation limit, the authors applied Newton’s Law of Cooling. The focus was on natural convection, assuming that because the container is suspended in seawater without external mechanical pumps for primary heat exchange, forced convection is negligible. Calculations for heat flux ($q$), total heat transfer power ($\phi$), and the convection heat transfer coefficient ($h$) were based on an aluminum alloy container with a thickness of 0.01m.
Key Insights and Results
- Orientation Matters: The study tested three orientations. Placing servers lengthwise (consistent with the container's length) was the most efficient, allowing for a physical maximum of 484 units. Placing them widthwise was the least efficient, allowing only 269 units.
- Heat is the Hard Limit: While the physical capacity ($n_1$) is 484, the thermal capacity ($n_2$) is calculated at 392 units. This confirms that in underwater data centers, the ability to shed heat is a stricter constraint than the physical space available.
- Final Design Recommendation: To ensure stable operation and maintain a maximum server operating temperature of 80°C, a single container shell should house no more than 392 1U servers.
Conclusion and Future Work
The model demonstrates that thermal dissipation is the primary bottleneck for seabed data center density. Future research could refine these findings by accounting for ocean currents (forced convection), which might significantly increase the heat transfer coefficient and allow for higher densities. Additionally, modeling internal forced convection from server fans could further optimize the thermal ceiling.
Summary updated