Morning everyone, 

 

I am looking to buy a new UPS from APC. I am under the impression that I can connect the UPS to my unRAID machine. The UPS will "safely" shut down unRAID, so that when the power does come back on, I do not need to do a 13 hour parity check. 

 

I am talking a look at APC UPS Back-UPS (BE850M2), what do you guys think? Will I need to buy any additional accessories? Any features I would be missing that would be helpful to have?

 

 

Hardware

• M/B: Gateway - RS780
• CPU: AMD Phenom™ 9100e Quad-Core @ 1800
• HVM: Enabled
• IOMMU: Disabled
• Cache: 512 kB, 2048 kB, 2048 kB
• Memory: 6 GB (max. installable capacity 8 GB)
• Network: bond0: fault-tolerance (active-backup), mtu 1500
•  eth0: 1000 Mb/s, full duplex, mtu 1500
• Kernel: Linux 4.14.16-unRAID x86_64
• OpenSSL: 1.0.2n
• Uptime: 5 days, 01:39:11 (5 days ago I had an outage)
• Flash device: Cruzer Blaze 8GB
• Cache: PNY CS1311 240GB SSD

• Parity	WDC_WD40EZRZ-00GXCB0 - 4 TB (sdd)
  Disk 1	WDC_WD40EZRZ-00GXCB0  - 4 TB (sde)
  Disk 2	WDC_WD40EZRZ-00WN9B0 - 4 TB (sdg)
  Disk 3	WDC_WD40EZRX-00SPEB0 - 4 TB (sdb)
  Disk 4	WDC_WD40EZRZ-00GXCB0 - 4 TB (sdc)

 

Edited May 15, 20188 yr by chris1bass
remove bad data

This will work for you.  However, a few things I would do if I were to use it.  First, I would set the "Time on battery before shutdown" to 30 seconds.  The first reason being that it is approximate sinewave voltage output.  This can be quite hard on the PS.  (Most modern PS will take this abuse for a short period without damage.)  Second reason is that it uses a rather small capacity battery for its max power output.    If you think about it, it is also quite practical.  If the power is out for 30 seconds, it will probably be out longer than the UPS can bridge anyway (~40m @ 100W). 

You do 30 seconds? I was thinking of leaving it on for a longer period of time before shutting down. At least a few minutes. I see that you have a cyberpower and apc UPS. any noticeable differences?

16 minutes ago, chris1bass said:

I was thinking of leaving it on for a longer period of time before shutting down.

 

Why? What is there to gain by doing so? Most temporary glitches are much shorter than that. As Frank said, if the power is off for 30 seconds it's very likely going to be off for longer than the UPS can maintain anyway. It's designed to give you enough time to perform an orderly shutdown, not to carry you through a power outage. The less you discharge the battery the shorter time it takes to recharge it.

It also covers the other problem that often occurs.   The power goes out.  The server shuts down successfully.    The power is out for an hour.  The UPS starts recharging its battery (which can require up to EIGHT hours).  After twenty minutes of up time, someone decides that  all is well.  Ten minutes later, the power goes back out.  Now, the UPS does not have enough charge on it to shut the server down.  You have an unclean shutdown...

 

Plus (as I said), the sinewave approximation voltage waveform is not that good for the server's PS.  (My PS's buzz when they run on it!!!)   If you want to run longer, you should be looking at the prue sinewave units. 

Edited May 16, 20188 yr by Frank1940

There's an interesting discussion here about whether modified sine waves are bad for switched mode power supplies.

11 hours ago, John_M said:

Why? What is there to gain by doing so? Most temporary glitches are much shorter than that.

 

I think that will vary quite a lot depending on where in the world you live.

 

Not just number of power failures/year but also average time for power to return will depend on what technology they use when implementing the grid.

 

I normally see maybe one-second failures for standard power grid failures before they automatically switch over to different distribution routes.

 

I normally see maybe minutes of failure if there is a local issue - which basically is someone very local performing some work and intentionally cutting off the power to block or building while quickly changing some hardware.

 

Once maybe every 5-10 years I see 10+ minutes long power failures and then there is often something very major that has failed requiring people to locate the fault and performing some on-site actions.

 

With a 5 minute timeout time, I might see equipment turned off once/5 years.

With a 30 second timeout time, I would probably see equipment turned off once/2 years.

Without any UPS at all, I would see the equipment lose power maybe once/year.

We will get about short hit about once or twice a month.  (The Digital clocks lose their minds!)  At what point these would give a problem to a server is a matter of conjecture.  These are the ones that I really worry about from my experiences from olden times.  What the problems seemed to be in those days is that things might not reset during a short outage and the system would come back up in an unstable state and maybe do a write with random (or corrupted) data to hard disk.  You would need the services of a witch doctor if the write was in the file allocation area.  Modern servers seem to handle this type of things more gracefully.  I can not recall the last time when the servers shut down due to a powerhit that lasted longer than 30 seconds that it was not out for a minimum of an hour in the last sixteen years.  I think eight hours was about the maximum.  (Luckily, we are only a mile from a major, major substation and the feed to the underground system used in our condo area to connected directly to one of the main distribution branches from that sub-station.  We tend to get restored quickly as the power company objective is restore power to as many people as possible to make them 'look good' to the regulators.  

31 minutes ago, Frank1940 said:

Modern servers seem to handle this type of things more gracefully.

 

All correctly designed microprocessor-based electronics today contains a power supervisor logic. Either a dedicated supervisor chip or some form of brownout detection in the processor. But it's standard that the processor gets reset long before the voltage drops to levels where the processor may start to perform random incorrect actions. Lots of embedded chips can also produce a brownout warning where the processor gets a number of milliseconds to quickly save important state information while still running on capacitors.

 

Some low-end gear still continues to use a RC-based reset, and that solution can't handle short glitches or voltages that ramps down or up.

 

I have had two power failures reaching 1 hour the last 20 years - one was a generator fire at a big power plant and one was replacement of a burned power cable to a neighbor building where the electricians powered off multiple buildings when installing the new cable. Where I live, hardly anyone uses any UPS - people figure they can live with one power failure/year that is often just a short blink.