[OpenIndiana-discuss] Powerloss protected SSDs for ...Re: Low low end server

[Jim Klimov]

On 2014-02-11 01:08, Alex Smith (K4RNT) wrote:
> Wouldn't a UPS with monitoring be a better alternative? Allow the server to
> power down safely on UPS when it detects a power lost state. Most UPSes I
> know have either serial or USB monitoring, sometimes even Ethernet on
> higher-end models (although I've never looked at the monitoring systems in
> depth, YMMV).

UPSes do also die, and their batteries in particular. Give a standard
model 3 years (5 to 10 on special models), and the UPS becomes a weird
multi-socket PDU :) And loss of wall power instantly translates to the
payloads. Also don't forget human error, wiggly cabling under a broom,
etc.

Luckily, many batteries are replaceable, and many are pretty standard
and cheaply available from OEMs, so renovation of an UPS with just a
new battery is cheaper than going off to buy a new one (as some do).

But this would still a matter of monitoring the UPS health state now ;)
And for an unmonitored location might be just as bad.

As for the SATA/SAS cable - I think beside a battery (or rechargeable
battery for that matter) a capacitor could as well be soldered on.
The real problem is informing the SSD firmware that it has just a
few seconds left to live and should evacuate data from RAM onto flash.

According to some woes (maybe on illumos/zfs lists, maybe on forums -
I don't remember exactly) there were "unprotected" SSD devices with
occasional problems even during standard shutdowns (which the UPS
solution may provide), in that the system flushed the disks and
turned its power source off as requested, but upon power-on the
SSD hardware could not find data or its own metadata - it was all
scrambled and garbled. YMMV of course, but such stories from the
trenches scared me enough to look for a piece of SSD which boasts
power-loss protection. At least, its engineers gave this problem
a thought some way or another :)

//Jim Klimov