Re: Status of Emdebian Grip

On 5/02/2014 8:36 PM, Neil Williams wrote:

>
> It is the amount of time I'm spending using current ARMv7 boards for
> work which is making Grip maintenance difficult. Grip needs developer
> time and I have better uses for that time when faced with a declining
> requirement.

I can certainly relate to that, Neil.  Time is my number one resource 
constraint also.

> BeagleXM is getting to be a very old bit of kit. Have you looked into
> the beaglebone (black or white)?

I'm just the hired help (software integrator), so I don't have much of a 
say in the hardware.  But if I were to put words in my Boss' mouth, I'd 
say that (a) new tech and high capacity RAM/flash costs too much, and 
(b) there are significant R&D lead times.   We are truly designing our 
own boards, and our programmers have been working hard to develop 
proprietary software to squeeze every last drop of performance out of 
the OMAP 3630's DSP; we're not just selling re-branded Debian and 
off-the-shelf devkits here.

The Beagleboard-based product is our established product line, we're 
already looking at TI's DM8148 for the next generation.

>> MiB of NAND flash in which to fit U-Boot, kernel, main userspace and
>> recovery userspace (yes, I know what you are going to say - this is a
>> "design flaw" that we brought on ourselves).
> It's a reliance on older hardware which may have a simpler fix.

No, mainly it's cost cutting.  The Beagleboard xM doesn't even have NAND 
flash.

The reason that thin client OEMs go out of business is the price of 
their units.  When large corporations have heavily discounted purchasing 
agreements with HP/Dell/whomever to bulk buy powerful modern PCs for the 
same or less than a far less capable thin client, then there's really no 
incentive for them to go thin.  So rather than selling works of art for 
 >$500, we're aiming to sell little black boxes for <$150.

I realise that BeagleBone already falls in the <$150 category, the above 
is just by way of motivational explanation for our aggressive 
cost-cutting, with reference to "traditional" thin clients.

> Emdebian Grip 3.1 isn't going away any time soon - the question here 
> is about unstable and testing migrations for Jessie. It looks, at the 
> moment, as if Emdebian Grip 4.0 Jessie is not going to happen. Given 
> that you have a relatively smooth way out to the upgraded version of 
> your current board from the same supplier, how important would point 
> releases of Emdebian Grip 3.x actually be? 

For me, 3.x point releases are really only important to ensure currency 
of data (e.g. timezone files, wireless regulatory DB, CA certificates, 
etc), not software.  Security vulnerabilities are a concern also, of 
course, but as this is generally not a customer-visible issue this is a 
mostly a risk we just accept rather than spend time mitigating.  
Appalling, I know, but that's the nature of competitive business!

Jessie-Grip, on the other hand, will be important for us.  Thin client 
distros (and having maintained my own one of these as an open source 
hobby since 2002 (DietPC, http://dietpc.org - too lightweight for 
ThinLinX purposes), I am speaking from experience here) are particularly 
sensitive to currency of [e]glibc and Xorg, and of course kernel version 
currency is always a background issue for everybody.

Wheezy-Grip is okay (barely) for our current product.  The official 
version of eglibc in Wheezy (2.13) is already too old for us, since 
Citrix has (to my annoyance) released Receiver for ARMHF with a higher 
glibc requirement; I had to substitute eglibc 2.17 from Sid. At the 
moment we're using proprietary TI and in-house code to compensate for 
the lack of open source support for certain OMAP features in the 3.0 
kernel we're forced to use, but for future products we could avoid some 
wheel reinvention by leveraging improvements in more recent kernel and 
Xorg releases, such as open sourced OMAP KMS DRM, PowerVR SGX 3D, and 
DSP bridge drivers.