General > General Technical Chat
Why aren't computers designed to handle power failure?
NiHaoMike:
--- Quote from: filssavi on June 12, 2020, 08:08:44 am ---- Backward compatibility: anything you do, if you want any hope of widespread adoption, it must be backward compatible with the ATX standard, mostly dating back to the 90's, so good luck.
- all PSUs now need to be intelligent enough to reliably detect a power loss and report it to the main processor, realistically the only way is coaxing intel into adding it to the ATX standard, because:
A) it needs to be universal, not a different communication method for eack PSU OEM
B) you actually need the OEMs to implement it in something other than their halo products, which is of questionable feasibility given the already razor thin margins in what ammounts to a commodity market
- Also software will probably need to be aware of emergency shutdown, as you only need to save the bare minimum to resume work (trying to save 16-32 GB to a bunch of spinning rust would realistically require way too much energy
- The solution will need to be batteryless (you can't expect a consumer to regularly replace batteries every couple of years, and beside good luck trying to find spare batteries for old motherboards
--- End quote ---
Just make it an ATX PSU with a standardized connection for a battery and USB for monitoring. In fact, with many modern ATX PSUs internally generating 5V and 3.3V from 12V using DC/DC converters and a 3S pack being just perfect for supplying 12V directly, it might not take much hacking to convert a PSU to do that.
What I would like to see is an additional signal that instantly drops the CPU speed to minimum to allow using a smaller battery, I have DIYed one fairly easily by adding a circuit to a motherboard that pulls down the PROCHOT line.
Monkeh:
--- Quote from: Raj on June 12, 2020, 07:28:40 am ---in America, where all the computers are designed
--- End quote ---
America must speak a lot of Chinese today.
madires:
Some statistics about power outages: https://www.indexmundi.com/facts/indicators/IC.ELC.OUTG/rankings
In my case about 5 short outages happen a year, mostly in the summer during thunderstorms. The outages last for a few seconds up to a few minutes (<10). Longer outages are quite rare (maybe one in 5 years). There are also a few undervoltage events.
bd139:
--- Quote from: pepelevamp on June 12, 2020, 12:39:01 pm ---** computer science guy joins the chat
the problem is that computers in principal should be able to turn off in the blink of an eye without any problem. yet we can't design general purpose computers without being drunk & smoking crack.
if you truly think about it - aside from say saving an open document - why would you really even need to do anything when you yank the power on a computer?
now i know you're all gonna bite at me saying oh gee pepe, filesystems need to be flushed to disk oh waa. to that i say - stop cheating to get more performance with ya wee DRAM memory caches.
anywhere that has a DRAM cache such as a disk must hold enough local power in a local supercap to flush it to disk.
the modern day CPU is full of lies & tricks to get more performance by cheating. writing to local caches etc. its all a big scam. pipe-lining, speculative execution, asynchronous volatile buffers - its all stuff that takes away your ability to deterministically run operations in sequence - all just so you can get more overall throughput. it makes it damn near impossible to yank out your power cable and plug it back in again & resume at point T: Your computer never has one point in time that its working at. Its scattered.
so I propose this:
either give every damn thing a supercap or stop lying to everybody & make honest computers.
it all falls apart anyway. look at speculative execution. the latest security screwup is so bad the only solution is to lock the entire memory bus & tank the system's performance by 97%.
--- End quote ---
Also comp sci guy. None of that really matters. All you need is a write barrier and transactional consistency. And we actually mostly have that.
In a lot of cases we have much much much better than that with journaling and nice new formats (Redis AOF I fucking love) which allow full ordered consistency.
Bud:
What is the point of connecting a battery directly to the desktop computer if it is often required to also keep the display powered and may be some other peripherals. A mains voltage UPS already does that.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version