[rescue] Sun T5240 power supply failure?

[Arno Kletzander]

Dave McGuire <mcguire at neurotica.com> wrote:
> On 03/03/2016 01:57 AM, Arno Kletzander wrote:
> > Just taking two Inlet Modules and sticking them in one chassis probably is a
> > REALLY STUPID idea (if not impossible because of the case panel), as AC mains 
> > is most probably just bussed internally between the two slots, meaning you're 
> > going to end up with exposed connector parts live, shorts if cables are plugged
> > into different circuits or backwards, and maybe other unfavourable events. It
> > would be imperative to implement some kind of isolation and failover logic to
> > make that work safely.
> 
>   Nah, those issues aren't really concerns in a real datacenter
> environment.  Someone making a smoke-releasing mistake like plugging
> them in "backwards" somehow (which is next to impossible to do) provides
> a good way to get rid of an incompetent employee in a critical environment.

Ahem? Deterrence in place of safeguards does not sounds like a good idea. There
is so much that can go wrong even without incompetence on the part of the person
actually installing the equipment. I know your mains sockets are polarized (ours
here in central Europe aren't! You can easily swap L and N by inverting the plug
and sometimes you even need to do that e.g. to get rid of mains hum in multi-
component Hi-Fi installations, or to stop flourescent lights from dimly glowing 
or flickering when off), but what's the chance of hitting a socket, extension or
IEC cord that somehow got wired up backwards? OK, you can buzz them out before-
hand, but there should be no need to do that for "consumer-installable equipment"
(basically anything with a mains plug/IEC inlet on it, as opposed to e.g. an elec-
trical kitchen stove that has a barrier strip and must be installed by a licensed 
electrician.)

Understand me right, I know this is not a problem with systems _intended and 
engineered_ for dual supply configuration (I have a couple of HP NetServers that
are), as I suppose those either won't care about the phase relation between the 
inputs (independent switchers only bussed on their DC side) or fail gracefully 
by not connecting to the faulty source and maybe lighting a telltale neon light
next to the IEC inlet. They also will not do dangerous stuff like letting mains
voltage appear on unused IEC inlets. The Sun E3000 chassis isn't anything like 
that. It has two adjacent slots for internally identical power supplies, one of
which has an IEC inlet (the PPS0) and the other one (PPS1) doesn't. AC power 
from the inlet is used by the PPS0, but also internally distributed to the PPS1
slot and used by that (if fitted - it is optional). 

Now imagine a system with two PPS0 modules installed, one of them already con-
nected to a power source, and somebody fumbling behind the machine to get an 
IEC connector seated in the other inlet. This one could easily turn into a 
"soul-releasing mistake"! (The supply modules do have switches on them which 
you could use to avoid that, but I don't think our code allows for that. Using
a micro switch that detects presence of an IEC plug in the inlets merely shifts
the problems to the other end of the cable.)

If the machine is connected to two different circuits, it would also "bridge"
between them in case one of them goes offline, trying to supply any consumers 
on the failed circuit and tripping the breaker on the second one as well if a
persistent fault condition exists. One more problem is that different circuits
can be powered by different phases if there is three-phase service. 

>   Multiple AC inputs are damn handy when "walking" a system between
> sources of power for maintenance or relocation purposes.  I myself have
> moved critical servers across town in a van without rebooting them.  One
> can't easily do stuff like that without multiple AC inputs.

Yeah, that's just the kind of stuff I was thinking about. I once witnessed our
University computing center staff load a server bound for some remote location,
alive and complete with bleeping rackmount UPS, in the back of a station wagon.
The other scenario is why one friend of mine isn't really sure whether a UPS is
actually a good idea, it introduces a new SPOF and is probably _less_ reliable
than just using mains directly in our part of the world. A dual-inlet machine
could use mains directly if the UPS gives up the ghost (but take care to use
a different circuit, in case the UPS trips the breaker), or you could even plug
it into _two_ UPSes for good measure.

>               -Dave
> 
> -- 
> Dave McGuire, AK4HZ
> New Kensington, PA

Arno