[AusNOG] Question about hardware spec for a DC

[Glen Turner]

Hi Mark,

> A few DCs (notably Nextgen/Metronode and Internode adl6) strongly recommend ATSs because there
> are OHS&W considerations which sometimes require isolation of switchboards when electricians are
> working behind the escutcheon.

I think those sites would prefer dual (or N+N) power supplies over ATSs 
where possible.

> But I can’t think of a good reason for not allowing ATSs, which, in most of the 1RU variants, are   
> basically equivalent to a piece of wire with a switch in it.

Exactly. So there is a risk that a ATS failure will bond the A feed to the 
B feed, dropping both. At least with dual power supplies a failure will 
only bond the low voltage outputs, a condition which is easily handled by 
the design of the power supply and both feeds will remain up.

I could perfectly understand a site which had a list of 'approved' ATSs 
where they felt comfortable with the design. Rather than allowing any old 
crap to threaten the integrity of both power feeds (especially if they 
have one pair of feeds serving multiple customers).

Feedthrough from the ATS into the 'dead' circuit is a OHAS issue. The 
correct response to reading that in the manual would be to install 
switches on the ATS inputs so that the feed and the ATS can be isolated 
before feed maintenance. This often needs to be done with DC rectifiers, 
as their capacitors can hold a hefty charge. It's simple enough: a 2RU 
blanking plate is just the right height to fasten industrial power 
switches. The labelling is vital, obviously; although I'd take care to 
make the layout very plain.

I'd be a tad annoyed at such an ATS not already coming with the switches 
built into the 1RU chassis, as it makes the installed height 2RU or so. 
Probably annoyed enough to cross them off the shortlist.

> There are lots of DCs which won’t let you run in-rack UPSs (they’re already providing you with UPS
> power, there aren’t a lot of reasons for the extra inefficiencies and power factor considerations involved
> in quadruple-conversion power, eh?).  

You'll recall a 15A UPS will pull 30A after it has been used (15A for the 
load and 15A to recharge the batteries). The site UPS then faces double 
the current draw than it otherwise would, for no additional protection for 
the customer. That doubling of site UPS capacity is alone enough reason 
for sites to ban downstream UPSs.

Battery fires within a computer room can be bad. There may be smoke and 
shrapnel damage to other customer's equipment, plus the damage from the 
fire fighting. This explains the popularity of distinct A and B UPS Rooms 
or Battery Rooms in high-availability computer centres. The hope is that a 
battery explosion or fire can be contained to the room and the main 
computer room see only a failure of one of the two power feeds.

> Fire regulations might require the DC to be able to shut equipment down before fire fighters start
> spraying water around.  They can’t do that if you have a UPS in your rack.

UPSs have an Emergency Power Off input. UPSs and other battery systems are 
required to have that input wired back to the room's EPO distribution 
relays. The room EPO itself can be controlled from the site's fire panel. 
This isn't a requirement peculiar to data centres; I am surprised by the 
number of comms closet UPSs which aren't cabled back to the building's 
fire panel.

Best wishes as always, glen