On 27/01/2021 16.39, Lew Wolfgang wrote:
On 1/27/21 3:35 AM, Carlos E. R. wrote:
A question about "battery": is it still an argument if you system is backed up by an UPS? I thought the batteries on controller cards were paramount only on systems that are not protected by a convenient UPS. Wrong? The idea is, I assume, that battery backed hardware raid detects that the power died and commits everything to disk before finally powering down the HDs.
The hardware RAID controllers that we use (3-ware/LSI/Broadcom) have super-capacitors instead of lithium batteries. The battery is used to power cache RAM that holds unwritten data in the event of a system crash. I don't think there's time for the controller to complete writes to disk, it doesn't power the disks after all.
Ah.
An UPS will keep the entire computer running till it finally gives up, or either the human or software powers it down before the battery runs down.
Not really the same thing, but both avoid disasters.
My experience with UPS may be limited, but back in the day I ran a 10-KVA UPS to power a group of Sun Microsystem servers. It was a nice UPS with a ferroresonant transformer that conditioned power as well as running the 10-KVA load for 30-minutes. In my experience, we experienced more unplanned power outages caused by the UPS itself than if we directly connected to the mains. After we retired the Suns and switched to SuSE I once managed to keep the main server up and running continuously for a bit more than 4-years, without the UPS. Without a reboot! It was a busy server too, with a hardware RAID controller.
Do you remember how did the UPS fail? I'm curious. I have a wild guess. I had once to install two sizable UPSs in cascade to power some "dumb terminals and printers". The instructions from upstream management was to put both in active mode, ie, directly and constantly power the terminals from battery, which were independently charged. This mode would age both boxes prematurely and make them fail eventually. So instead I connected (at least one, maybe the two, I don't remember) in passive mode: ie, the terminals were supplied from mains, and fail over fast to battery on power failure. I told my boss later, when I saw him, he agreed. The only advantage I see with their method was that the AC supplied to the terminals would be under our control. But a good UPS will switch over if the mains is bad, either low or high voltage or frequency change, and do it so fast that a computer doesn't crash. -- Cheers / Saludos, Carlos E. R. (from 15.2 x86_64 at Telcontar)