[AusNOG] What's Typical Power Density in Colo?

Mark Foster mark.foster at smxemail.com
Fri May 4 10:07:31 EST 2012

On 04/05/12 02:43, Julien Goodwin wrote:
> On 03/05/12 20:43, jason andrade wrote:
>>> Must be N+1 (ala fully redundant power) and have reasonable connectivity.
>> Which one did you want ? N+1 or redundant (2N) or fully redundant (2N+2N)..
> Ugh, power redundancy.
> So there's:
> * 1 - Single feed, no redundancy
> * N - One or multiple feeds, no dependency
> * N+1 - One or multiple feeds, handles any single feed failure
> * 2N - One or multiple feeds, handles 50% failure
> * N+2 - One or multiple feeds, handles any two feeds failing
> * 2N+1 - One or multiple feeds, handles 50% failure, and one more feed
> failure
> And that's just off the top of my head.
> Of course there's all the fun cases where power *feeds* don't equal
> power *supplies*, which is the case on some stuff (A bunch of the large
> Juniper's just to start). Then there's the whole load-sharing vs.
> failover cases (can your breakers actually handle it if you lose a feed,
> what about starting your systems from cold)
> You always want to be *very* clear what you mean, especially as some kit
> can be configured in multiple ways (even in software, eg HP bladesystem).
> The traditional telco way to do this was two main building feeds, A and
> B, and you order an X amp feed, or an X amp feed A+B if you want redundancy.

Yeah, it's a tough world. 
One Datacentre I have a bit of experience with used an approach that
seemed fairly solid:

- A+B Power on a per-rack basis (eg two seperate 16A feeds.  They were
both on the same phase, which initially seemed counter-intuitive (what
if you lose the red phase?) but apparently there's good reasons to do
- Power to the rack distribution board balanced across multiple UPSs
(N+1 - the failure of one UPS would not cause demand to exceed UPS
capacity and push you into UPS bypass)
- Power to the UPS's supplied by direct mains, switched option to
Generator, either of which was sufficient to cope with the demand of the
UPSs. (N+1 Power, i.e. two options to feed power to your UPS's, being 1
more than the 1 required ordinarily).
- Note that transition from Mains to Generators obviously requires the
UPS's to be within spec... UPS bypass would cause gear to die during a
transition-to-Generator.  (The whole point of the exercise is the fact
you don't have much influence on the stability of your commercial mains
supply, right?)

2N (double the requirements across the board) can be a little overboard
IMHO, unless you're particularly sensitive to the probability that you
will lose more than one of any critical component at any one time. Thus
why I usually target N+1 where I can.   For a typical footprint I would
guess you'd want:

- N+1 PSU equipment (redundancy in the case of PSU failure)
- A+B Rack Power Supply (redundancy in the case of PDU failure or supply
- N+1 UPS Supply (redundancy in the case of a single UPS failure)
- N+1 Mains Supply (redundancy in the case of your mains power failing)

An alternative model i've seen used, particularly for smaller / less
capital intensive installations, is:

- N+1 PSU equipment
- A + B PDUs
- UPS feeding PDU A
- Direct Mains feeding PDU B.

In the event of mains failure your load swings to the UPS side. In the
event of UPS failure you still have mains.

Obviously in all of the above you need to manage your load to be less
than the capacity of N, or you put yourself at severe risk.    So in the
case of my blade chassis, which has 4 PSU's and requires 2 to be
supplied to work, i'd expect to split my
load 2-by-2 across two PDUs in the knowledge that any one of those PDUs
will carry my operational load without blowing up.

(And if I'm feeling particularly sensitive about my non-dual-supply
equipment, one of those supply-switched PDUs with all my single-supply
gear hanging off it, would seem to be the go.)


This email has been filtered by SMX. For more information visit smxemail.com

More information about the AusNOG mailing list