[Arm-netbook] EOMA server standard

[Gordan Bobic]

On 10/26/2012 05:07 PM, luke.leighton wrote:
> On Fri, Oct 26, 2012 at 4:43 PM, Gordan Bobic<gordan at bobich.net>  wrote:
>
>>>    no - that was referring to the existing EOMA-68 standard, which is a
>>> standard designed for mass-volume appliances, *not* servers.  EOMA-68
>>> just happens to include USB3, 10/100/1000 Ethernet and SATA-III.
>>
>> So you are saying a EOMA68 "desktop" card can't be used in an EOMA68
>> "server" chassis?
>
>   that's incorrect.  ok - it's assuming that there *is* an standard
> which re-uses PCMCIA for server modular purposes.
>
>> That's a serious limitation. What can be done to make
>> them compatible on the overlapping pins/features?
>
>   i thought long and hard about this, 18+ months ago, and i couldn't
> think of a good way that didn't place a significant burden on the
> electronics (cards, server chassis or appliance chassis).   so i
> decided to focus on appliances first, and leave servers until later,
> maybe find a different connector or something.
>
>>>    yes.  but in doing so, you now force *all* devices to have *exactly*
>>> those features.  they can NOT be optional.  so if the standard
>>> requires 2x 10/100/1000 Ethernet, then any SoCs which do not have 2x
>>> 10/100/1000 Ethernet are automatically excluded.  likewise for any
>>> other features.
>>
>> Why can the features not be optional? Why is it necessary for the card
>> to support 2 ethernet ports just because the chassis can handle it? Why
>> not simply leave those pins disconnected?
>
>   i've been over this a number of times, but let's go over it again,
> can you please double-check and see if i missed anything?
>
>   ok, let's say that you have a chassis with 2 ethernet ports and 2
> SATA drives.  one is the boot drive, the other the data drive, and
> it's for a router system.
>
>   card blows up: you rush out, buy another one, put it in and FUCK the
> FUCKING thing doesn't boot, what the hell's wrong with this thing??

Standard doesn't mean there are no optional features. Backward/forward 
compatibility is far more valuable than all the features being the same. 
Some chassies will require different features from others, and thus not 
all EOMA cards will be suitable for use with all the chassies. That 
doesn't mean there is no standard.

PCs are extremely standardised. But that doesn't mean I can assume that 
the GPU is on-board. It might be, it might not be. If I don't want to 
buy a separate graphics card, I have to make sure that the motherboard 
has it built in.

This does not detract from having standards with optional extras.

>   turns out you bought a card that only has *one* SATA and *one*
> ethernet: it won't boot, and not only that, but when you swap the
> drives over in the chassis to find this out, fuck me if it doesn't
> route the damn traffic because there's no 2nd ethernet on the card.

You'd be in the same boat if you were swapping ATX motherboards. If you 
aren't paying attention to what you are buying you deserve everything 
you're going to get. The case you are describing isn't really 
reasonable, and is a bit like buying an AMD CPU motherboard and then 
moaning that your Intel CPU doesn't fit into the socket.

>   i'm using swear-words because if i was in a data centre and i bought
> a replacement card and it didn't work as advertised and my job was on
> the line, i would *definitely* be swearing :)

If you went and bought an incompatible card before checking the 
requirements people should be swearing at you. :)

>   and i would be blaming that stupid, stupid "EOMA server" standard for
> having done something as damn stupid as to make it possible for me to
> get confused over not reading the words on the box at the store, "only
> has 1 ethernet and 1 SATA".

Heavens forbid that people should actually check the spec on what they 
are buying before they actually buy it for a specific purpose. That's 
just crazy talk, right? </sarcasm>

The way I see the EOMA standard is more like the ATX standard in the 
sense that the core pin assignments are the same, but that doesn't mean 
that some pins shouldn't be used in different ways depending on which 
optional features are used.

>   so rather than let that happen, the rule is: there *are* no optional
> ports.  you can have "upgradeable functionality" e.g. offering 8 wires
> for up to 1000 ethernet with the option either side to
> down-level-negotiate to only 4, but definitely *not* "1 port is
> entirely optional".

If you want to mandate the "one size fits nobody" solution, then great, 
but you can't then moan that it isn't a standard suitable for every 
eventuality. Your solution to this seems to be having multiple standards 
for multiple use-cases, which is arguably worse than having one standard 
with optional features. All you are doing is fragmenting your standard 
into multiple incompatible standards, as opposed to having one standard 
where different features are optional.

>   this is something that was drummed in to me at university, from the
> lesson our lecturer taught us about the X-25 standard.

Don't take everything you're taught at university (or anywhere else) for 
the absolute truth. It is in human nature to "teach" others our 
opinions, rather than objective facts, because people's perception of 
truth differs. And then there is also the enormous gap between the 
academic idealism and real world pragmatism that has to be bridged 
outside of academia where we actually have to make stuff that works and 
is workable rather than some idealistic approximation that is completely 
unimplementable.

Gordan