Ian Seaton, CPI’s Technical Applications Development Manager, recently received the following question.
Question: What are your recommendations on fan controls and the best points to use for driving a feedback loop?
I have been reading quite a bit about Chatsworth for the past five months, and now we have an opportunity to install the cabinets, which is very exciting to be perfectly honest, but I wanted to run something by you.
I was thinking about the controls to implement for this project, I know from reading when there is a raised floor, the strategy can be as simple as pressure transducers located in the floor to monitor the pressure, and adjust the ECM motors to deliver a constant pressure to all areas. So if there is a movement in overall pressure, then the fan will adjust and lower the output, but keep a constant pressure. That could be one strategy, which is similar to how a variable volume terminal can work as well.
But in our case, we have a room that we are blowing the air into, so I have been thinking about temperature and if that is a possible way to control the fans. Let's say the fan servers start to shut or wind down, less air required, but I do not think there will be a significant amount of temperature drop in the ducted return, at least not enough to warrant sending a control signal to a fan VFD and change the speed. Because I think by the time there is a change in temperature, say even 1° F degree, this will take some time. Then there is a signal to the fan to modulate the CFM, okay, now it takes another few minutes for the reduced airflow to bring the temperature back up to 90° F (i.e. return set point). I just don't think that it's worth making an adjustment to the fan speed based on minor changes in the system. There is probably a threshold where the change is warranted, but the more I think about it, the more I think that changing CFM across the fan to account for small changes in server fan operations is not worth the change. I think of it in terms of what level of sensitivity do we want to employ. It's similar to cycling a modulating valve, something you want to avoid, as the system over and under shoots the set point. I was wondering if my view is in the ball park and if Chatsworth has had some experience with using temperature as the control point. I could see using it, but maybe the sensitivity is based on every 2° F change or some value. I am inclined to use temperature, but just make the fan change for some value, maybe 2° F degrees, to allow the entire system to reach some steady state condition and make it's own adjustments first, before changing set points.
So then I started thinking about adding pressure transducers to the room, only it's not a controlled environment like under a raised floor with perforated tiles, it's more like a room full of eddy currents and cross flows, and I do not see a transducer providing much semblance of accuracy that I would want to use for a control signal.
So I move to the other option, adding pressure transducers to measure the differential across the server cabinets. I think of it again, as strategies that can be used with variable terminals, where the pressure is maintained only to keep open the most demanding terminal. So if all the server cabinets drop off in fan use, then the differential pressure with the biggest requirement is satisfied, but this is not a controlled environment like the inside of a duct, and the pressures are not really the same as those required for a variable terminal system, because the DP across the cabinet is only one component of the total static pressure required, not the whole as with vav system. I do not see DP sensors across the cabinets being a good control measure for fan VFD control, but maybe I am missing something here.
I would greatly appreciate your help and recommendations.
Answer: This might be the first time my answer is shorter than the question.
I think you have done a pretty good job in articulating the reasons why you would not want to pursue various feedback loops. Your understanding is quite good in those areas. The bottom line is that you don’t want to get into a situation of micromanaging the airflow into the data center. Let’s just consider for a moment the relationship between airflow and load and temperature rise. This relationship is described by the equation CFM = 3.1W ÷ ΔT. If you had a small space with 20 cabinets with four HP c7000 blade centers in each cabinet, the total airflow demand in the room would be 34,640, and then you would add a factor to that for natural bypass. We usually assume 20%, but in a new, well designed space that can actually be as little as 5%. For the sake of argument, let’s just use 10%, so in this room, you would be delivering about 38,100 CFM. So let’s say you turned off a server and removed that demand from the equation. That would raise your excess supply from 10% to 11% and it would probably just turn into bypass. Even if that just didn’t become additional bypass and it was actually “forced” through the servers, the result would be a 0.3° F reduction in the temperature rise through each server due to the extra air supplied, based on that sensible cooling relationship equation. That excess supply would equate to less than $2000 over a 12 month period, based on running 4 air handlers for an N+1 redundancy at 75% capacity and thereby consuming 42% of the nominal energy requirement (.753 = .42), which would be reduced to 37% at the lower airflow demand created by one server shutting down. That equates to a 17520 annual kW/h difference, or $1752 @ $0.10 per kW/h. In a larger data center that difference would be significantly smaller and in a smaller data center that difference would be larger, but it gives you some sense of proportion for what you want to invest in controlling the air delivery.
So, all that being said, what are the best points to use for driving a feedback loop? Measuring the pressure of the room is still a viable alternative, but you need to consider the issues raised because you don’t want the pressure metric contaminated by air velocities in the room. The key is to find a “dead zone” in the room where the effect of air velocity is minimal. A CFD model can help locate likely dead zones. In this scenario, you are just looking at one or two sensors to measure what we call “gross static pressure.”
Again, you want to allow a certain amount of variation so you’re not micromanaging this space, which can conceivably put you into a cycle of continuous adjustment because the room never has a chance to stabilize. Power might be another metric point for driving a feedback loop. After all, the relationship between air flow, load and temperature rise is fixed. Once you have the room correctly determined, then output changes to the UPS are going to reflect air flow demand changes in the room. Again, remember the sensitivity described in the above example for shutting down one blade center and be sure to allow some tolerance of variation.
Find out more about CPI Passive Cooling. For additional help call us at 800-834-4969 or leave a comment. Ian Seaton, Technical Applications Development Manager