Datacenter simulation
 

Hi there...

I would like to model a datacenter.

At this moment I consider the problem as follow:

- Rectangular shape
- 22 racks are considered as 1 volume with constant heat source (W/m^3)
- The air density is considered as constant
- Assumption is that there is a laminar flow
- The simulations are steady-state and all the parameters are considered as first-order in time and space (pressure-based solver)

Can someone give me some tips about the BC's which are normally used to model the racks in a datacenter? Does it need to contain all the racks...1 row contains 22 racks and each rack again contains 60 vertically placed IT-equipment.

I would like to hear what you think about this settings...how complex must the model become to have a reliable model? Because I only want to know whether all the equipments are in contact with the cooling air and what the temperature difference becomes over the equipment.

Thanks

SolarX,

What software package are you using? I am interested in doing the same thing, but am looking for a decent open-source package to use.

Thanks,

Mark

Hi Mark,

I'm using Ansys Fluent!

What is your problem about?

Regards

> I'm using Ansys Fluent!

I did a search on data center and neglected to see where you had posted the the message. Whoops :-)

> What is your problem about?

I want to do some basic thermal modeling of the rack (not the whole room). We want to experiment with load balancing on the servers to maximize cooling efficiency.

This work is on a limited budget so I wanted to see if there was a recommended open source CFD package that people could recommend.

Mark

It is important to be as accurate as possible when doing CFD analysis. If you can avoid making an assumption dont make it.

To answer your questions:

1) there are no open source CFD programs. These programs are extremely expensive to develop. I currently pay 55k annual for on user in 6 sigma. I used to use ansys airpak and paid 35k annual for that. You can get tileflow for 16k annual. I switched because 6sigma by far gives me the most accurate results. It eliminates assumptions, and helps improve the geometery of the racks, as well as allows me model airlfow on a/server basis instead of on a rack basis like the other programs.

2) For geometry assumptions, you can make them in corners far away from AC units and servers. Say within 15 feet of an AC unit or server do yuor best to make sure everything is accurate to +/-1 or 2 inches. Watch for things like gaps where there should or shouldnt be gaps.

3) Assuming 1 block of racks to act as 22 racks won't do you that much good. You need to account for things like blanking panels that aren't installed, and as a minimum get load correct on a per rack bassis(can get you results of +/-5F @ 90% of points). I recommend per server basis but this can be time consuming(can get you results of +/-2F @ 90% of points). The difference is huge if you consider that ASHRAES recommended thermal envelope is only 12F.

4) Steady state parameters are fine. As long as you can figure out how to measure steady state when validating the model:) this takes lots of practice. Most CFD programs automatically accoutn for density changes. Some CFD programs allow you to choose turbulent instead of laminar. IF your program does I suggest you take advantage of this option. The airflow comming out of AC units will almost always be turbulent.

Hopefully this help gets you started. I do about 1 CFD model a month for various clients and I love to give advice. If you have more questions fire away.

for CFD modeling garbage in = garbage out so if you can avoid making an assumption do it.

PS. I am quite busy and don't check these boards often so if it takes awhile for me to respond fire me off an email and I'll respond on the board for you.