[Arm-netbook] EOMA server standard

[luke.leighton]

On Fri, Oct 26, 2012 at 12:29 AM, Baybal Ni <nikulinpi at gmail.com> wrote:
> What do you think about sticking with "twisted pair on PCB" then? It's
> the same conventional twisted pair ethernet, except for no transformer
> couplers and it uses tracks on PCB instead of wires. This way we are
> getting free from "different phy" problem. Integrators will then only
> have to choose between 1000T(8 tracks) and 100X(4 tracks)

 that's what EOMA-68 does right now.  then, if a chassis or a CPU Card
only has 4 tracks, auto-negotiation takes care of down-negotiation to
100mbit, and if a chassis or a CPU Card has only 8 tracks,
auto-negotiation takes care of selecting 1000T.

 this isn't my main concern, however.  my main concerns are: a) 10G or
b) how and whether to do multiple lanes of 1000T.

 if multiple lanes of 1000T are specified, then because of the reasons
already explained (nothing in a standard can be optional) then *EVERY*
server-based CPU Card *MUST* provide multiple lanes of 1000T.

 in cases where a particular SoC doesn't have multiple-lane 1000T?
tough shit - they've *got* to provide that exact number of lanes, by
e.g. putting in a PCIe bridge PHY and a 1000T PCIe PHY IC, or a
USB3-to-1000T PHY IC etc. etc. qty as specified in the standard.

 10GBase-T at least does auto-negotation over the exact same 8 wires.
it just uses 2.4 watts to do so, which is horrendous.

tough call, i know.

l.