[133085] in North American Network Operators' Group
Re: Want to move to all 208V for server racks
daemon@ATHENA.MIT.EDU (Joel Jaeggli)
Fri Dec 3 23:00:16 2010
In-Reply-To: <AANLkTim1w=9kBSSVOsJeaUv4riPOSeM_KfPqPtFhFrPe@mail.gmail.com>
From: Joel Jaeggli <joelja@bogus.com>
Date: Fri, 3 Dec 2010 20:01:48 -0800
To: Matthew Petach <mpetach@netflight.com>
Cc: "nanog@nanog.org" <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Dec 3, 2010, at 19:25, Matthew Petach <mpetach@netflight.com> wrote:
> On Fri, Dec 3, 2010 at 7:18 PM, Joel Jaeggli <joelja@bogus.com> wrote:
>> On Dec 3, 2010, at 16:58, Valdis.Kletnieks@vt.edu wrote:
>>=20
>> On Fri, 03 Dec 2010 13:21:07 PST, Matthew Petach said:
>>=20
>> People are still feeding their gear with AC? Save on PS inefficiency,
>>=20
>> and feed direct 12/5vDC to the servers. Save space, save power,
>>=20
>> save cooling.
>>=20
>> What does that do to customer equipment choices? I've got a quarter acre=
of
>> boxes that I know want 12/5vDC inside the case, but that's not an easily
>> available option from the vendor - most of the time the only option is
>> autoswitching 120-240DC with your choice of power cables.
>>=20
>> The 10,000amp bus for the 12v feed for a row of server racks would be a
>> thing to behold. I don't think anyone but Paul Wall has seriously conside=
red
>> this.
>=20
> Some day I'd love to meet that guy--he sure has come up with some
> revolutionary ideas here!
>=20
> (OK, so it's not as practical when you have other customers to worry
> about... but it might not be so crazy when you're looking at the
> efficiency numbers for 100,000 small 1u power supplies vs a set
> of much larger ones.)
Ohm's law is a bitch. 10kamp -48v DC plants are bad enough as far as the amo=
unt of copper required, running 12v for significant distance is comical, thi=
s is the reason small boats airplanes and diesel trucks adopt 24v systems. T=
here's probably some model where top of rack rectifiers makes sense but that=
's really pretty much what a blade server is. When you look at a motherboard=
in a server a big chunk of of real-estate is devoted to taking 12v and swit=
ching it down to 1.2-1.8 for distribution to the CPU/memory, a 4 socket serv=
er might have to carry 400amp around in a space of around 300cm^2 on a layer=
of the pcb.=20
The justification for running 208 or 480 all the way to a cabinet is all abo=
ut smaller conductors.
Joel
>=20
> Matt
>=20
>=20
