I seem to remember installing an HPE switch recently that had fan modules that could be reversed. You're absolutely right -- this is an obvious and sensible solution to a self-inflicted problem.
FWIW, I don't think software-reversible fans are a realistic option, because of the points made above about fan blade shape and the engineering related to that. But as long as the fan is spinning the correct direction, the switch is already engineered for the airflow to be push or pull. Turning the whole module around would accomplish that without affecting the performance of the fan itself.
The fans already need a mechanism to report speed, so an i2c link or even a pin that is either high or low depending on which way it was installed in the chassis, would report back the orientation and ensure all the modules are in agreement. If not, you would get an alarm and the fans would spin at full RPM to alert you that something is wrong. Thus preventing a mixed-up push/pull arrangement from compromising thermals.
The PSU is a little more difficult, since the fan is usually fixed, and space is tight enough to make a reversible module somewhat unwieldy. (This is what we get for asking for 1000W in 1U.) One solution is to move the cooling out of the PSU module itself, and use air channels to the adjacent fans. But, since PSUs are often OEM'd from a dedicated manufacture specializing in PSU design and safety certification, perhaps that's one best left alone. I would take having to swap the PSUs rather than having to swap two PSUs and five fan modules for completely different SKUs. That really is just kind of nuts.
Original Message:
Sent: 04-15-2021 03:17
From: Unknown User
Subject: An opportunity to Engineer Simplicity
I think the concept is a great one. However one would need to consider re-engineering the fans so that they will support bi-directional flow. The airfoils on fans today have been specifically engineered for best efficiency, maximizing air volume and maintaining a somewhat quiet level of noise at a set speed and direction. Making a bi-directional fan will have to sacrifice some of this efficiency and noise levels will increase significantly.
If we can overcome these limitations, I believe this would be adopted by all manufacturers.
There will need to be some provision for preventing misconfiguration as well. This could not be something incorporated as a configuration option but more in line with port channelization and VC setup groups.
Original Message:
Sent: 04-08-2021 10:36
From: Unknown User
Subject: An opportunity to Engineer Simplicity
We have an issue trying to manage our switch inventory against air-flow needs. We have a good number of switches in network closets what are typically AFO. We have a good number of switches in Data Centers which may be AFO or AFI depending on how they are mounted in the racks which typically depends on the orientation of network ports on other equipment. Managing airflow orientation of inventory is a real pain.
Here is an opportunity for Juniper to Engineer for Simplicity.
Simply install DC fans in fan and power assemblies. Add direction control to the speed control already provided for fans. Add an LED that will show blue (AFI (cold)), or orange (AFO (hot)) to indicate operating air-flow direction. Add a configuration element such as: set chassis air-flow <airflow-direction> or perhaps set chassis member <member-id> air-flow <air-flow-direction> or perhaps set virtual-chassis member <member-id> air-flow <air-flow-direction>.
Presto-changeo airflow direction becomes irrelevant to manufacturing, logistics, inventory, ordering, deployment.
It seems to me that this would eliminate the complexity of managing air-flow orientation for both Juniper and its customers.
This seems so simple and obvious, yet Junipers newest switch family (EX4400) has fixed air-flow orientation.