Microsoft had a Windows Azure container on display on the 2009 PDC. The attendees could take a look inside and experience the cooling in the container. Every now and then, the doors of the container
were closed, and the cooling was switched on so that the people inside could
experience the cooling. When I was inside I could feel a very strong and cold
wind flowing from one side to the other side of the container. The airflow was
much stronger than what you'd expect.
The container uses an adiabatic cooling system. A garden hose provides water to a
spongelike wall. This is not water cooling like what you know from the high-end
gaming systems, where the water is used to directly cool the CPU coolers. With
the adiabatic cooling system, the water is just used to increase the humidity
in the container. The wind blowing through the spongelike wet structure
evaporates the water, which decreases the temperature and increases the
humidity. Apparently the server hardware is designed in such a way that it will
keep working reliably at a humidity of 80%, but Microsoft would like to go even
higher. It's up to the vendor of the container to design the hardware in such a
way that it supports such humidity. The wind is generated by huge fans at the
exhaust, and generates an underpressure that results in rather high wind speeds.
Pictures (not the best quality, these pictures were taken with a cell phone):
 Outside view of the Azure container. The big metal
vents allow the intake of fresh air. This container model is an
outside model, you can just put it as-is on a parking lot. |
 Here
you can see the connection of the container to the power grid and internet. |
Closer view of the power and internet connection |
Power and internet entering the container. |
To the left: adiabatic cooling wall. To the right: the azure hardware. |
Extremely densely populated racks. |
Close up of the sponge-like adiabatic wall. |
|
This container was apparently built by Dell. |
|
Microsoft has explained in the past that these containers are left untouched until sufficiently enough nodes have failed. At this point, the container would be returned
to the manufacturer to be serviced. I asked why it would then be needed to provide space for
human intervention. Apparently the policy of shipping the container back to the
manufacturer is made on a project by project basis.