My automation system, while highly distributed, is, for the most part, powered from a central location. I.e., a PoE switch acts to deliver power to most of the field (I think there is *one* node that has "local power").
The servers are colo'd with the switch. Probably 60%? of the system's power is consumed in the equipment closet. The balance is distributed around the field (but, again, *delivered* from the switch in the equipment closet).
Everything *in* the equipment closet is essentially +5 & +12V -- once you get *beyond* the line operated power supplies. The field is powered from 48VDC (i.e., PoE).
The prototype system is powered by a pair of 1500VA COTS UPS's. Management functions in the switch allow me to dynamically shed load to make the best of the power available from the UPS's (e.g., the weather station doesn't need to be powered during an outage!)
Ignoring, for the moment, the PoE needs... it seems like a smarter solution would be to replace the power supplies *in* the various servers and other equipment closet contents with a single (big)
+5/+12V supply. And, further, power that supply directly from
*battery* -- charging the battery from line current (when available) much like the POTS. Hopefully, a *single* line operated supply ("charger") can more effectively power the complete load than a bunch of smaller power supplies, each powering a *portion* of the load (off the COTS UPS!). And, hopefully the conversion efficiency DC-to-DC would be better than DC-to-AC-then-back-to-DC (via the COTS UPS).
The pisser is the 48VDC for the switch! If it was "just a small portion" of the load, then I could see treating it as an exception (i.e., design for the "local load" and treat the 48V as "accessory").
But, given that it represents a sizeable portion of the load, it seems like it will have a more pronounced influence on the overall design and its operating efficiency. E.g., powering a bunch of +5/+12V loads "from a battery" leads to a different choice of battery (voltage, etc.) than having to also accommodate this higher voltage component of the system.
What should I use for back-of-napkin guidance towards which approach to take (high battery voltage oriented towards the 48V supply's needs; or lower battery voltage oriented towards the +5/+12V loads)?
And, comments as to how this approach would compare to the COTS COTS-UPS-feeding-COTS-power_supplies?
All of these still put a switcher in between your primary battery and your devices, but...
I know you can get power supplies that have a 9-15 V DC (or so) input and standard ATX outputs, designed for running a PC in a vehicle.
I also know that you can buy a power supply in the standard ATX size, with standard ATX outputs, but that takes -48 V DC input, for running a PC from a "telecom" power plant.
I *think* you can buy power supplies in the "server" sizes (1U or other oddballs) that have standard outputs and take -48 V DC input, but these are likely to be uncheap.
If you want to be real adventurous, a lot of AC-input switching power supplies will *work* if you feed them at least a couple of hundred volts DC. Using them this way is not on the label, though. This may not buy you that much in terms of efficiency, but maybe it simplifies the design of the battery side.
You can also get DC-DC converter bricks that have 48 V inputs and whatever you want for the outputs. If you standardize on a 48 V battery, maybe these would work as PC power supplies. You'd have to generate the "power OK" signal to the motherboard, and handle the "turn on" request line from the motherboard.
If you haven't already, I'd get in there with a Kill-A-Watt or two, and see how much (AC) power is actually being used by various parts of your system. This will tell you if your PoE switch is sucking most of the juice, or if it's the PCs, or what. You might have to figure out how to make a bunch of the PoE devices active at once, to get a good idea of how much the switch draws.
If the Kill-A-Watt can handle the output (frequency/waveform) of the UPS, it'd be interesting to instrument both the AC output and the battery on the UPS, to find out how efficient it is.
Since you're probably going to be making exactly one of these, I'd say to prowl the surplus places, eBay, etc for cheap new or used gear, and see if that guides you towards 12 V or 48 V battery. The 48 V telecom stuff is nice and will last a long time, but is probably uneconomical to buy new for your house.
You will probably have a little bit of battery distribution in your server room. Lower voltage means thicker cables, but it also means that all the fuses, switches, etc that you might need are as close as your local car parts store. 48 V lets you use thinner wire, but you have to hunt around a little more for the parts.
If you have a COTS UPS and COTS power supplies, and one of them pops, a replacement is available... well... OTS. :) If you cook up something on your own, replacement parts are your baby.
If one of the server motherboards crowbars its +5 V or +12 V rail, the
300 W (or whatever) power supply for that server will just quit. (It might shut down or it might pop a fuse, but the damage should be limited to that one server.) If you have multiple servers fed from one big 1500 W supply, a fault in one server might take down more than one machine. Also, the big power supply might dump juice into the fault for a while before it shuts down.
Say that the combined efficiency of all the power supplies in the chain you have now is 80%, and you think you can get that to 90% with your proposed scheme. If what you have now draws 2000 W from the wall plug, the proposed scheme would draw 1780 W - a savings of 5.28 kWh a day, or
1930 kWh a year. If electricity is fifteen cents a kilowatt hour, this saves $290 a year. (Subtract from savings for running the furnace more in the winter and add to savings for running the A/C less in the summer.) You can decide how soon you want it to pay back, and figure out how much you can spend on it.
The PoE switch current consumption depends on the number and type of the PoE devices connected to it, so the specified switch current requirement is the worst case situation. Check your actual loads connected to the PoE switch.
+5 V and +3.3 V needs to be regulated anyway and +12 V doesn't usually need that amount of regulation, so a +12 V battery feed might be usable.
Of course 48 or 60 V are usable, but since these are "telecom" voltages, the equipment seems to be quite expensive.
24 V PC power supplies are usually much cheaper and used widely in automation systems.
Yes, but the PoE switch is already a potential single point of failure (as it distributes power to MOST of the loads)
Yes. But it's *one* switcher instead of one in the inverter to get
*up* to 120VAC and then another(s) in the "power suppli(es) to get back *down* to 5/12 (and 48)
The goal is to eventually remove all but, perhaps, one "PC" (i.e., "truly high performance device acting AT a bottleneck" -- the sort of thing it would be foolish to try to DESIGN/BUILD as its required features are so "generic")
I've been steadily increasing the capabilities of the "field" components and migrating performance *out* of the servers. This allows the size of the "server closet" to shrink (I'm now down to a couple of cubic feet -- mostly for the switch, UPS's and cabling) as well as moving power dissipation to more "distributed" locations (so ventilation in the closet doesn't then become a problem). It also gives me a lot more *useable* MIPS.
I.e., any "COTS PC" would be the *exception* in the power budget, not the "rule".
I've been making *small* increases in the power consumed in the field (in terms of watts for computational) while making *big* reductions in the power consumed in the servers. E.g., I believe I can now fully deploy with less than 1.5KW. I'd like to trim this by an additional factor of 3-5 for "nominal" operation (realizing that you can always be piggish in how you use what's "available").
I try to actively manage ALL resources (in the environment AND in the system) so a user can, theoretically, *cap* the cost and let the system's performance adapt to that cap (over short and long intervals).
E.g., if you want to limit power consumed, if you're burning X watts delivering music to your bedroom speakers, the system can defer burning Y watts in the irrigation system (on the assumption that an opportunity will arise when those Y watts will be available... because you turned off the music *or* some other load has been shed).
Sort of like not running the dishwasher if you're planning on taking a long, hot shower! :>
The UPS efficiencies vary with load. But, pretty close to ~90%. Of course, simply having the UPS *there* makes the system less efficient than it needs to be (no free lunch).
No. My goal is to come up with a design that others can use "as is". If, for example, I required several COTS servers in the implementation, then this makes deployment impractical for most folks -- too much space required, too much heat to tuck it all away in a closet, etc.
Bigger UPS's like to run at 48VDC (lower battery currents for a given load). But, more expensive battery! (i.e., you're buying them "four at a time" -- and, they are a maintenance item)
Given that I have a lot of control over the load(s), I am trying to come up with the best approach for making power available -- knowing that "backup" is a key issue for at least some portions of the system.
[E.g., the alarm system has its own dedicated battery so that even if the switch dies, it remains operational "for a very long time" to minimize the use of building power as an attack vector.]
I could, for example, design the switch to be powered off 12-24V and have little ~15W converters at each PoE port. Or, a *big* converter that feeds the entire switch with 5/12/48VDC. I could also run a pair of "primary DC power" busses through the device to allow "line cards" to tap into one *or* the other (i.e., this port will or will not be capable of battery backup operation)
[ISTR sending some of this to you in a PM -- but can't find a record of that. Maybe faulty "meatware" :-/ (Or, it is one of the accounts that I closed :< ]
The 48V PoE requirement is "unfortunate" -- given that the loads at the end of the drops regulate this down to 3/5V -- but understandable given the power delivery requirements and the cable capacity carrying it.
That's the case, regardless. If someone fries a network device, you can't go buy the PSE *or* PD component OTS. Reliability always has an associated cost (hot/cold spares, etc.) -- the UPS is one such obvious "cost"! I think the wiser effort is to shrink the *needs* to a point where these events are less likely *and* less costly!
E.g., how does Joe Casual User evaluate/understand the consequences of putting N PC's in a small closed space? Does he know what the reliability consequences are likely to be? ("Gee, my overclocked PC isn't working correctly. I wonder *WHY*...") The ubiquity of things like PC's have led to a (incorrect) assumption by many that they "know all about them". As if they were kitchen blenders or mixers (yet people still manage to "burn out" even these *simple* kitchen appliances by misusing them!). To some extent, a computing "appliance" can avoid/defer that attitude because it *isn't* so "familiar".
You *know* you need a "spare tire" for your vehicle because you know that you can't repurpose something *else* to fit that need WHEN it arises!
*I* can spend quite a lot! :> But, that doesn't mean others would want to make that same dollar tradeoff. :< E.g., I deliberately chose a very constrained location for the "server closet" (the lowest shelf in the pantry!) instead of a more "favorable" location (the store *room*) as I felt my choice would be more in line with the resources folks would be able to "make available". A set of 1U servers in a short rack would be *easier* ("COTS") to acquire. But, deployment (space, cooling, noise) would be problematic.
[I catch a lot of grief when I have *my* servers running for a few hours at a time; I expect I'd be living in The Dog House if I developed a system that required something *similar* to be running 24/7/365! :-( ]
It's also not "responsible" design practices to come up with inefficient solutions when efficient alternatives exist (but you didn't EXPLORE them; hence my post!)
I closely control the "instantaneous" power demanded by PD's/delivered by the PSE. It's not a naive "negotiate-a-class-at-connection-time" approach from the standards.
E.g., I may opt to use the processor in a network speaker as a *compute* resource and power down the attached audio amplifier. So, the node no longer needs the power that the amplifier *might* deliver to the attached speaker(s).
I don't think I would count on the 12V from a battery to power any nominal 12V loads. E.g., I'm not sure most such loads would tolerate a battery at 13-14VDC... *or*, down around *9*!
But, a converter (12V-to-12V) operating at 12V can use cheaper/smaller semiconductors than a power supply operating at 120VAC.
This makes sense. In your original post, though, you mentioned the need to power at least one server from the proposed UPS, which is why I was thinking of ways to rig a "PC" power supply with a DC input.
Well, run it on just one 1500 VA UPS and see what happens! :) (Yes, I know VA != W).
I once pencilled out a UPS that had a weight on a rope. Use a motor to lift the weight to charge; use the rope to turn a generator to discharge. It turned out not to be practical for the electrical requirement (125 W for about 4 hours) in the space available.
That, and it matches telecom DC plant voltage, which originally needed to shove juice through a few miles of tiny copper.
You sent me an email about your plan to protect the Ethernet lines from somebody shooting RICH, CHUNKY VOLTS into them, but not about your DC UPS plans.
I am trying to get that to be little more than a COTS disk drive with the "compute" portion moved out into the field. (COTS disk wouldn't fare well in the locations where most of the other nodes are located.
Yes. But also makes it harder (more expensive) for others to replicate. E.g., powering everything off 12V (or 24V) batteries would allow for a cheaper solution. In my *recent* experience, power outages are a rarity. Aside from the occasional "flicker" (which seems to happen once or twice a week), there are very few prolonged outages, here.
OTOH, when I lived in Chicagoland, there was one per month...
So, how inclined will folks be to invest monies in batteries if they are seldom/rarely required?
Ah. Sorry, I couldn't find the "sent" message so couldn't recall its content. End of the year I do a fair bit of "disk cleaning" :-/
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.