Comment by bob1029 | Hacker News Reader

Comment by bob1029 | Hacker News Reader

Over the years I have lost a lot of confidence in UPS units which do not use double conversion topology.

I have a fairly expensive line-interactive unit that can be tricked into doing bad things just by power cycling my laser printer which is plugged into the same circuit (but not the actual UPS). One day this caused me more trouble than it was worth WRT lost work. I would have been better off plugged directly into the grid and just taking the 100 millisecond voltage dip at the PC's power supply. This is one area where buying high quality PSU can go a long way. High wattage Seasonic units can ride through some pretty nasty conditions before needing any outside help.

I also have a cheaper passive-standby modified-sinewave unit which has never misbehaved in terms of trigger conditions, but it is hot, noisy and otherwise does not inspire any confidence when it's running on its inverter. I would also never dream of plugging anything into this unit that uses a non-switching power supply.

Again, I am faced with the reality that you get precisely what you pay for. I always assumed double conversion was expensive paranoia never for home use, but over time this has proven out just like everything else when looking at the value equation of technology.

I too use double conversion UPS. Liebert Vertiv GXT4 in my case. The power in my neighborhood is sloppy and the power company refuse to replace the transformer until it goes out entirely. The neighbor has a septic pump that creates nasty spikes. I've had multiple APC UPS literally catch on fire. It's always the tower models with the real time voltage regulators that catch on fire. The floor power strip models have been rock solid, but they would beep every time the neighbors pump kicked on. I tried power conditioners, filters, isolation transformers, but in the end, it was the double conversion UPS that cleaned everything up and the Liebert has provided reliable power ever since. Some day I want to go solar, but I would have to move somewhere else first.

> I've had multiple APC UPS literally catch on fire. It's always the tower models with the real time voltage regulators that catch on fire.

I wonder how that is even possible in the first place.

Harmonic distortion and back-emf?

Total speculation.

Man. Ive got like 3 of those suckers all of them with fairly decent usage. Been needing to swap them out. Two of them are laying on their side stacked.

Probably is a fire hazard

> just by power cycling my laser printer which is plugged into the same circuit

So I encountered in a previous IT helpdesk job a laser printer connected to one of the tower APC units, with a wattage display.

Laser printer (it was a Ricoh Aficio SP 4100-something MICR printer for checks IIRC) was connected to the battery backup set of receptacles for some odd reason.

Any time the laser printer would start up, or run, and I presume charge the fuser, it would briefly spike up to 900w and often set off an alarm. It was rated for 550w. I think sudden high loads like that could be damaging to the unit.

I would not attempt to connect a laser printer to a UPS at all, even if not connected to backup battery--IMHO it's in the same category as high-current devices like toasters and hair dryers. Many commercial copiers I've seen have some type of large surge protector but no UPS.

Laser printers plugged into UPS's aren't great (and then there was my previous boss who destroyed a pretty large unit by repeatedly boiling a kettle from it during a power cut), but it's worth pointing out that the OP had his laser plugged into the same circuit as the UPS but not onto the battery - I took that to mean it's on the same radial/ring/breaker rather than connected to the UPS at all.

I'm guessing the surge current to the laser was causing the UPS to register a brownout and potentially not behave correctly at that point.

My Cyberpower 1500VA actually says not to plug laser printers into the Battery Load side. It's rated for 1500W, but you know it won't work when the manufacturer says it.

My own UPS system is focused around keeping the telephone system and one DSL modem up and running through 2 days without power. That way if there is a bad thunderstorm there is no panic about working at home, I can use a laptop and tablets and communicate as much as I need.

I have a PC tower server and realized off the bat that it would not be feasible to supply it with backup power for any worthwhile period of time. If the power goes out it will crash, but it runs on ext4 and ZFS and odds are very good it will come back when the power does and put itself back together.

I’m with you on zfs (although it is a pain to restore overwritten metadata, that’s not a typical symptom of power/controller failure) but please change ext4 to jfs or xfs. Xfs has a bad rep from the early naughts in terms of fs corruption but actually addressed the underlying issues and today both those options are equally great.

I’ve tested all three extensively with unreliable media (raspberry pi kernel dev testing on an sd card) and ext4 was the absolute worst.

Can you elaborate for which storage mediums ext4 was the worst? The microSD cards? Or even normal SSDs?

All. Even in the tests where the SD cards never physically failed, their different write patterns and flushing guarantees exposed underlying issues in the file system. You just don’t encounter them as often when working on systems that typically align with the assumptions the fs developers incorrectly made.

Quite interesting.

What is the best FS for microSD cards in your experience?

NILFS2 or F2FS, the latter is much more heavily tested as it is now the default on a number of extremely popular Android phones.

SD cards don’t perform wear leveling so you really need a file system that minimizes/distributes block rewrites, and a log-structured filesystem will do that by never rewriting old data (until there is no space left).

I don't need days of power, but I wanted a few hours. I ended up getting a Belkin Battery Backup REV B, BU3DC001-12V off of ebay and modified the barrel connector to fit my cable modem and router. I can get at least 3 hours (I haven't tested it longer than that). I'll probably get another while they're still available.

As for my server, I have it on a UPS big enough to allow it to shut down (with some slack for battery degradation). That's all I really need.

If you want to run a handful of low current DC devices for an eternity and in a rock-solid manner, just get a DC-DC converter for each and a few batteries+charger. You can keep the devices permanently looped in on the battery while it is on trickle charge. Think of it as a car's electrical system in your network cabinet, replacing the alternator with grid power.

If we are just talking about cable modem + wifi router + NAS, you cant be looking at more than 30 watts. Buck/boost DC-DC converters are upwards of 98% efficient, so you don't have a bunch of losses with that extraneous inverter/rectifier stage. All of this stuff can be purchased and assembled with minimal fuss.

That's why I went with the Belkin, it isn't a "normal" UPS. 120V comes in and is stepped down to 12V, but there's no up-conversion the other way; the output is 12V (up to 3A) to a barrel connector. I bought a Y so I could energize my two devices (after making sure combined they don't exceed 3A).

Is there a term for UPS supplies for this purpose?

I want one for work to keep the network and internet devices running, but it needs about 30 Watts continuously for say 24 hours, not hundreds of Watts for 30 minutes like a UPS for a server.

I presume a "trickle" power UPS would need to care about power efficiency?

It shouldn't matter, just remember to shut off the alarm for power savings.

You should also consider about SSD's reliability without PLP capacitor. NAND flash is erased by block and it runs GC silently so possibly even non-touched data could be corrupted if power loss.

I wonder that most consumer SSDs are aimed to power loss safe (by implementing like CoW?) even without capacitor but I don't know the truth.

> I wonder that most consumer SSDs are aimed to power loss safe (by implementing like CoW?) even without capacitor but I don't know the truth.

The flash translation layer in a SSD looks a lot like a journalling or log-structured filesystem, and without power loss protection capacitors you get similar data integrity guarantees: unexpected power loss may mean loss of recently written data (still in the drive's write buffers, and maybe stuff in partially-written NAND pages), but shouldn't cause catastrophic loss of data. You generally have more data at risk from being buffered in main system RAM by the OS than you do from the volatile/unprotected buffers in the SSD itself.

The SSD should never be erasing a block as part of a garbage collection process until the live data from that block has been committed to a new block.

Thanks. Your explaination makes sense but I want to hear such things from SSD manufacturer.

Reply on news.ycombinator.com