# How to measure airflow going through individual server vents

*Sorry everyone for the very wordy question.

I am trying to write up a method to accurately measure the airflow through the inlet and various outlets on one of our servers in order to measure the heat output from the server partitions/cards. The boss wants to know the rate of heat production from each card, the current airflow provided to each card, and what the airflow actually should be to keep the cards running at a certain temperature. The server is partitioned into three sections and the outlet vents look similar to these:

I have purchased a TSI 9565 velocity meter that comes with a hot wire anemometer and a small pitot tube that measures down to 0.15m/s. I plan on using the pitot tube attachment to measure the air velocity, but my concern is that the velocity will vary depending on where on the vent I take it from, so I won't just be able to average it. The problem server is on site and the technician has limited time to access the server, which means he will only be able to take one measurement at each vent. We do have an exact copy of the server setup here in the main office, but the rack is quite different which means the airflow going into each server will be different.

However, I was thinking that because the server setups are the same, maybe the ratios between the velocities for each vent would also be the same. I don't have any justification for this, but if it is roughly true, I could record each individual velocity reading from each vent perforation on the server here in the office, where I have plenty of time to do it. Then I could specific a certain spot on each vent for the technician on site to take a single velocity reading from and find the factor of difference between that server and the office server. From there I could multiply all the velocity values for that vent that I recorded on the office server. It's a bit hard to explain so I've included a quick example.

My question is, is my assumption that the servers would have the same relative velocity profiles valid? If not, what would be an accurate way to measure the airflow through each vent?

• Possible duplicate of How to accurately measure mass flow through server vents using an anemometer – Solar Mike Nov 29 '17 at 3:45
• Yes I could not delete the question before getting an answer. I didn't write the original question clearly enough – david_10001 Nov 29 '17 at 6:25
• I think you will need a different strategy : something like a tube that connect to each vent that allows a turbulent non-uniform air flow to develop at least some sort of profile then take the measurement across the end of the tube... – Solar Mike Nov 29 '17 at 8:34
• @SolarMike - The first question reads to me to be more about "what tool and technique should I use for Foo?" And this question appears to be "how do I specify tool use for Foo?" I think it would help both of these questions if david_10001 edit this question and clarified the difference with the previous question. – user16 Nov 29 '17 at 12:54

"Is my assumption that the servers would have the same relative velocity profiles valid?" Assuming your server's output air temperature is $343~K$ max. And from the experimental data, max velocity is about ~$3~m/s$ also assuming the individual port of ~$5~mm$ in size.

So in the field server, the Reynold's number ($Re$) is approximately ~$10^4$ . and in office (same temperature & size but the velocity is 1.5 m/s) ($Re$) < 10000. both are well within in the laminar region (it is only the exit condition).

Since the velocity distribution is caused by same geometry(in field & office) and Reynold's no indicates that the regime of flow are ~same (laminar regime), assuming linearity between two of your measurements might work. (for velocity profiles & mass flow rate only).

Linking from the previous question, you are intended to find mass flow rate & heat thrown out from the server.

When you come to heat transfer calculations, you have to measure the temperature profile for both the servers in the same way you did for velocity profiles.

Temperature profiles not necessarily similar because, total heat transfer is depends on local heat transfer coefficients (mainly convection by air) which are non linear function of local Nusselt number & local Re number. So it may not bring the similar temperature profile at the outlet.

So the flows with similar velocity profiles may not have similar temperature profiles.

Hope this helps.

• 243K max is what temperature in Degrees C? Is it that cold?? – Solar Mike Nov 29 '17 at 12:44
• Most data centers are kept in the range of 20-24 C, and a server could easily be pushing 60-70 C at the CPU. Given the challenges they have had with this particular server, I wouldn't be surprised to hear if it was getting closer to 100 C – user16 Nov 29 '17 at 13:02
• @GlenH7 so 20 C is 293K as far as I remember so would you think 243K is just a bit low.... – Solar Mike Nov 29 '17 at 13:16
• Sorry .. edited now – mustang Nov 29 '17 at 13:29
• Thanks for your answer, it helps a lot. For the temperature readings, we are setting up some loggers at each of the outlets on site. I originally figured the heat distribution across each vent would be fairly even - I'm not sure if this is a valid assumption though. Would you recommend I carry out the same process for the temperatures as I did for the velocities i.e. measure each square on the office server? Or could I assume that temperature distribution across the vents would be fairly even? – david_10001 Nov 29 '17 at 22:14