For a while I have been thinking about centrally powering all of my devices in a specific communication or server closet. I'm currently using power adapters that came with devices, and that solution does work well. Given that essentially every MT device that I know has a power in Port maybe it would be more practical to utilize PoE switch to feed them all? In my head that looks like an interesting and practical way to restart any of the devices externally if need be. Additionally, this could be a good way to have an administrative only interface on those devices.However, I am not sure what is the best practice of doing so. The downside of this would be losing one of the parts for power, and I am sure there are other things that I haven't think of.

I hope someone could share their own experiences and practices regarding this idea of mine because I'm sure some of you do this.

Thanks!

In theory powering centrally is a good idea. One can use a high-quality power adapter with high efficiency for example.

When it comes to practice, with Mikrotik things become a bit complicated ... all devices support passive PoE, but not all devices can use same voltage range. Only some devices support standard IEEE 802.3 af/at PoE ... but standard PoE gives flexibility of using other vendors' PSE (PoE switches) if so desired. Additionally, most af/at PoE switches are actively cooled (noise) which may not be welcome if central location is anywhere near living space. A side note: 802.3 af/at uses 48V (up to 57V by standard).

So I'd start with inventory of what kind of PoE is needed for devices (802.3 af/at vs. passive, voltages). If there's a PoE state which covers all (most) of devices, then it's worth doing it.

what mkx said, it's usually worth doing it. There's active / passive poe converters available if necessary too but as being said before: Check what kind of PoE the to be powered device actually accepts.

I personally don't PoE my MT gear but I've implemented a setup with a PoE switch that powers Hue bridges via Poe -> Power Jack converter. Same story here: Jack has to physically fit the device and voltage / amps requirements on the output side must meet the demand of the powered device.

Thanks

My general idea was to have a CRS328 or CRS112 in a rack, and then power the devices through it. Both of these switches have dual voltage output where 328 as PSU-s integrated and 112 via external adapters. I would skip non MT solutions and injecting power, because that would defeat the purpose in my view (also, 328 switches are really reliable - 112 not so). The current practice I have with adapters was a redundancy solution, where it is less likely that all adapters that are fed over UPS will die so I just replace the ones that died. However, remote restart when device hangs up is a problem this way.

So say we have a rack: Would something like this be a good setup?

The idea to connect / loose the first port for general usage is valid? - I handle my devices in star configuration and prefer that more powerful router routes and handles all traffic and physical links.

Additionally, is it sensible to leave dual power in some cases to the device? Adapter + PoE for redundancy (altho I am not sure which one takes precedence)?

Heck, I could even route WAN connection from ISP router to the internal router, power all of the devices, and also have separate management backup VPN to manage devices directly specifically is something is broken inside the network itself... (It woudl be nice to have a switch with physical bypass ports like on RB1100AH, so if I use dual power and 328 dies I can still have internet and entire network would work, only I would receive an error.)

Well, once your devices are on the roof / high on the wall you can either add a power plug close or run network cables only and be done.

Adapter does take precedence over PoE and I'm not sure how Mikrotik switches go about this but it's not uninterrupted either so switches will at least do a power cycle in that scenario. If power redundancy is of concern I'd have 2 UPS with bypass panels in that rack so even if you have to change batteries / USV your PoE Switch would be online. I am unsure if you can hotplug a new PSU into a 328, if possible I'd provide power exclusively over the 328. If that's not possible I'd take a different switch with hot pluggable PSUs.

Adapter does take precedence over PoE ...

Not really. Precedence has power source with higher voltage (because devices internally feature simple circuitry with diodes blocking power bleeding).

Not really. Precedence has power source with higher voltage (because devices internally feature simple circuitry with diodes blocking power bleeding).

ok, great. Passive Poe Input range for CRS326 is 10 - 30V, jack input is 10 - 30V so assuming both CRS328 and the power supply go with 26V for input, where's the precedence now?

Mikrotik supports the Ohm's Law "standard" (e.g. Watts = Volts * Amps)

Re: precedence on INPUT power, multiple power source should be fine but it NOT be a "good idea" to mix voltages of the power supplies (but likely work). Now precedence on OUTPUT PoE is generally software controllable on most switches (via "PoE priority").

I think you need to lookup at the max power consumption of the device & the max output power of the PoE port for the devices involved first. And finally you need to make sure the total of all ports power is within the limits of the PoE switch. Three things to check before using passive PoE (e.g. typically ~24V). The 802 at/af/etc standard (e.g. typically 48V) neogiciates all this, so it just won't turn on if the power requirements aren't met (but still has some limit on max power per port and capacity, which is what's used by standard to determine if power is allowed).

These are all in Mikrotik's spec pages for all of there devices... But doing the "power math" is kinda important with passive PoE, so that what determines whether is "good idea to PoE feed all ROS".

If we pick a small PoE switch for example, like the RB260GSP, that has a max of 1A output per port. And limit of 2A for all PoE output. The "watt output" is determined by the input voltage to the switch for passive PoE. Let's take your 26V supply, so max watts per port would be 26 watts (1A * 26V), with max of 52 watts (2A * 26V). You'd need to size the power supply to include the power needed by switch. (so for RB260GSP add 5W for it's own power, or 0.2 amps (200mA) at 26V).

For the CRS326 says max of 21 watts. And let's take your 26V supply. So if you used the 5.5/2.1mm jack without PoE, you'd need at least 0.8 amp (800mA) of power. Now the issue with PoE is there is loss over line, so while you may start with 26V, it voltage will drop over the line, so that same 26V from switch to router over ethernet will be lower at the end of the ethernet line. This depends on cable quality and length, so let pick it's 22V at the end of 100m run, now you'd need to start with a power supply of at least 1.0 amp (1000mA) at the starting switch. But this is why you see a range listed on the acceptable voltage: voltage drops with cable length, so the device deals with voltage it's getting (as long as in range), thus starting at higher end of voltage supported at start is important.

well my question was: Which power Input takes priority on a Switch that has both PoE In and a Power Supply connected to it as I still do believe if actual mesured voltage on both inputs would be the same that would defeat "simple circuitry" because if the cable between the CRS328 and a CRS326 is 1m losses are very small (which is in a rack not unlikely) for one and secondly if that happens it would blow "simple circuitry" to hell and FCC wouldn't approve that.

However I'm no electrical engineer, I am to this point still sure the switch would take it's power from the power adapter rather than PoE-In, if power adapter fails as a source of power it would do a reboot aswell and not take PoE-In as "hot standby" power source either.

Passive Poe Input range for CRS326 is 10 - 30V, jack input is 10 - 30V so assuming both CRS328 and the power supply go with 26V for input, where's the precedence now?

If both voltages are exactly the same (at the point after those diodes), then device will draw power from both sources ... exact power ratio will depend on voltage drop after current flows (so resistance of wires, internal redistance of PA, ... do matter). But more likely one of PAs with same nominal rating will provide slightly higher voltage (usual IT PAs aren't precission voltage regulated) and tgat PA will then provide all (or most if voltage drop is big enough) power.

multiple power source should be fine but it NOT be a "good idea" to mix voltages of the power supplies

Why do you see mixing of supply voltages as a problem?

well my question was: Which power Input takes priority on a Switch that has both PoE In and a Power Supply connected to it

For switch own consumption, it'll be the highest voltage. For PoE out it'll be power from PA. A short while ago some MT staffer explained, that PoE cascading doesn't work in most multi-port switches, it only works on small devices (e.g. cAP with only 2 ethernet ports, one is PoE in, the other one is PoE out).
And it's logical, every PoE has power limitations. If it's passive, then PoE in will likely be limited by power source and highest limit I've seen has RBGPOE injector with 2A limit, most PoE switches have far less (e.g. 600mA). 2A at 24V gives 48W and that's not really a large power budget to power many attached PoE devices. If PoE is standard 802.3 af/at, then there are various power classes, but for af/at they are all less than 30W. 802.3 bt brings us higher power budgets, but MT doesn't support it (yet). Again, that power budget (on PoE in) doesn't really allow for many PoE out ports.

It shouldn't reboot if they're close, at least that's the way it works on a lot of the Mikrotik's we use. But I can't speak to the CRS line since we don't use them.

I normally don't do this in the real world, but tested this on a wAPacR, which has a wide voltage range. If I use 12V on the jack and 48V 802at switch, it uses the switch (based on /system/health showing 51V). If I remove the PoE, it doesn't reboot, and reads 12V (since at this point that's the only power). NOW... Only when I add back the PoE at 48V (e.g. volts going UP), does it reboot. After the reboot with both 12V DC and ~48V "enterprise" PoE 51V, it goes back to 51V. So, at least on wAPacR, it only reboot if voltages goes up on the PoE.

yes, I'm trying to keep it simple: CRS328 1m cable length will provide 26V on POE-In, Power supply on CRS326 will provide 26V in to CRS326, no PoE-out there as that wasn't stated by OP either. According to you: Highest voltage takes precedence, in this scenario both voltages are exactly the same, where does the CRS326 take it's power from? I am aware this isn't a common scenario but with passive PoE and the specs of the CRS326 it's possible.

I normally don't do this in the real world, but tested this on a wAPacR, which has a wide voltage range. If I use 12V on the jack and 48V 802at switch, ...

Interesting. Did you try the same, but uding supplied 24V power brick as "alternative" power source? 12V is almost at lowest allowed and it could affect the whole thing.

.. in this scenario both voltages are exactly the same, where does the CRS326 take it's power from?

Again: it's not likely that voltages will be exactly the same. As @Amm0 noted, hus 48V power adapter actually supplies 51V (give or take). But if both supplies really get at same voltage to CRS326, then power will be drawn from both. Probably not 50:50 because directly attached PA with its thick wires will gave less resistance and voltage drop will thus be less (at same amperage) than via PoE. The ratio will be such that voltage drop due to resistances will be same for both power sources.

You have to understand that powering Mikrotiks from multiple sources is not binary (on/off), it's analogue.

I normally don't do this in the real world, but tested this on a wAPacR, which has a wide voltage range. If I use 12V on the jack and 48V 802at switch, ...

Interesting. Did you try the same, but uding supplied 24V power brick as "alternative" power source? 12V is almost at lowest allowed and it could affect the whole thing.

I kinda picked an extreme (12V passive + 802.3at from Netgear PoE switch). And certainly believe the different hardware has different levels of sophistication in the power electronics, so YMMV...

But Mikrotik-supplied 24V adapter on DC, with same 802.3at switch, on same wAPacR, did same thing: Reboot when going running on 24V passive, and enterprise/802.3at 48V PoE is plugged in. Could be some delay negotiate on 802.3at on this one too...

Now if I use passive PoE at 24V using the Mikrotik PoE injector, and 24V via the 5.5/2.1 DC jack at same time. Both two Mikrotik AC/DC power supplies, one to PoE inject and one directly plugged in to wAP. it doesn't reboot regardless of how you remove/add power, as long as one is there.

What happens with passive 48V PoE with 12-24V DC via jack was left untried (but may try at some point but don't have an switches that do that here).

I'm aware it's not binary, VRMs rarely are binary I can accept it might actually draw power from both sources at same voltage supplied and to me it kinda makes sense that in the case of the wAPacR it would reboot after active PoE comes back as it would be design wise a better choice to prefer the higher voltage supply as it has to handle less amps there on feed in and even makes sense that it handles a voltage drop in a standby scenario without reboot but not a voltage increase back to normal operation.

Rebooting device when transitioning from local power source to 802.3 af/at PoE is unfortunate behaviour. Did anybody submit a support ticket about it?

Transition between voltages should not reboot device. Devices have internal (and autonomous) DC-DC downconverters (CPUs typically run on 1-3V, another downconversion is needed for USB port if device has got one). And DC-DC downconverters regulate output voltage a few thousand times per second, output voltage is smoothed using capacitors ... so change in input voltage in any direction should not reboot device (if capacitors are at their specs).

Rebooting device when transitioning from local power source to 802.3 af/at PoE is unfortunate behaviour. Did anybody submit a support ticket about it?

I haven't. I can see that being a use case on the wAP, just not one I've run into. But was kinda surprised it rebooted, tried a few times too, seemed repo'able. I'll try 48V passive at some point, and also try disconnecting power at the switch (vs un/plugging the ethernet jack, which may introduce more variability as the cable is removed/added etc.). But we provide 24V via PoE and DC-in on these (and older RB9xx devices too), and don't recall reboots if one is removed. Now CRS specifics I just dunno, other usual Mikrotik deals with "minor" voltage variations without issue.

To the original question, "Is it good idea to PoE feed all ROS and SWOS devices centrally in a closet?"
My answer is still be Yes – just with footnotes.

Agreed, it's good to feed all devices centrally if possible but will take some preparation on the rack, too. If power outages are a big concern even more so and an UPS won't run without maintenance forever either / might still die so I personally wouldn't want to miss a bypass panel or better two with 2 UPS so there won't be any tears if the only UPS actually decides to stretch it's legs for the last time.

Another option for power cycling switches without being onsite is PDUs that allow for cutting power to outlets on it, I'd get a RackPDU that supports that and eventually if desired for the switches placed somewhere else smaller units, too.

I'd recommend using a 24V lithium battery as the UPS (Amazon, etc has a few if you search for "24V dc battery pack"), vs some office UPS that uses AC and lead-acid.