Data Center expansion is a multi-variate undertaking that can lead to unforeseen constraints and unnecessary expenditure if underlying assumptions are not confronted. Often intuition is so strong that certain elements are assumed to be a given, but determining how and whether the expansion will fit within all elements of the existing infrastructure takes careful analysis. A cooling system’s capability can fall into this category, as there are often multiple solutions to any expansion project.
Consider the following example as a case study in miniature – the addition of two more racks to an existing Coolant Distribution Unit (CDU) supporting a bank of four Rear Door Heat Exchangers (RDHXs). The chill water supply is 40°F, and the return is set at 60°F. The one-inch tubing inside the CDU connects to a ¾-inch manifold supplying 62°F water to the rear doors with 74°F returning through the return manifold. The rear door heat exchangers are required to pull 10kW from the exhaust air of each cabinet. As the chill water system and the CDU have ample thermal capacity, the manifold inlet valve is set at 45% open and the individual control valves for the heat exchangers have all been manually set to the 70% open.
Here is a model of the system in operation.
The first attempt may be to open all the control valves to compensate for the additional load, especially since the Manifold Inlet Control Valve is halfway closed, and we’re only increasing the thermal load by 50%. Setting the manifold inlet control at 95% and the individual rack control valves to 90% provides the overall 60kW cooling required at the CDU. However, now only two of the six racks are getting the 10kW they need (colors showing a relative degree of cooling shortfall).
We can try to manually balance through a number of trials, but that is tedious and requires multiple iterations. With the new thermal control feature in PIPE-FLO® Professional, we can set the control valves to provide the flow required to each rack to cool 10kW. Doing this we find a problem – we have exceeded the capability of the control valves for the two added units and one of the previously existing units. How can we solve this?
There may be a temptation to upgrade the CDU to a higher thermal capacity or change out the control valves, but are we jumping to conclusions? Using the Gradient Color feature we can look at the hydraulic bottlenecks – areas of high head loss – to help determine the best course of action.
At first glance it appears the key elements are the control valves in the CDU, but if we narrow the range for head loss to see only pipe sections (0-5 ft) it reveals we have a significant loss in the first section of both the supply and return manifold headers.
This leads us to some alternatives. The manifold system appears undersized for 6 RDHXs, and we could replace the ¾-inch manifolds with 1-inch tubing as shown below, which reduces the head loss and easily achieves the 10kW per rack desired.
However, an even lower cost alternative is instead of connecting the new loops to the far end of the headers; we change where the supply lines attach to the headers. We can achieve this by either expanding the header in the other direction or moving the supply lines to a different node (replace a tee with a cross) as shown below. Now we can add the two racks with minimal modification to the system.
While this is a small example, PIPE-FLO® Professional can perform similar analyses on your system, and there is no limit on the number of pipes or devices. With PIPE-FLO® Professional, even large-scale projects can be manageable and efficient.