[Server-devel] "Gigabyte BRIX more scalable than Intel NUC"

[James Cameron]

On Mon, Sep 05, 2016 at 10:25:09AM -0400, Adam Holt wrote:
> On Terry's recommendation to look at replaceable WiFi modules that
> can serve "almost 50 kids", my current interest is to explore the
> $279.99 Gigabyte BRIX GB-BSi3H-6100:
> 
> http://www.gigabyte.com/products/product-page.aspx?pid=5691

Looks good.

> If we go with the above CPU/chassis -- or anything similar folks
> recommend -- which WiFi module (PCIe M.2 presumably) to consider for
> maximum community support isn't an easy question of course: any
> recommendations for Linux support?

At OLPC in the past months I've been testing two M.2 cards and two
MiniPCI Express cards.  Not the special ones for XO-4, but industry
standard ones.  And not for access point mode, just as client.
Realtek is the chipset vendor.  Don't know if the cards are available
retail, as we're focused on the volume side of things.  M.2 is
certainly a more forward looking connector design to choose from;
higher density, more bus features.

I'm worried about your approach.  Don't fixate on product selection?
While it might seem to be the only thing you can control, your problem
is affected by a whole stack of components; wireless module,
motherboard host controller, kernel driver for host controller, kernel
driver for wireless module, kernel driver for access point mode,
programs for access point mode, and more.  I'd draw you a picture but
my crayon is hiding.

> Bluetooth 4.2 Low Energy support would be an Optional Bonus, as we
> have a long-term desire to provide teacher smartphones' full control
> over their "personal" school server.

Bad idea.  Bluetooth LE will use the same radiofrequency spectrum as
the wireless access point.  Co-existence means switching between the
two modes, and that switching will deafen one or the other mode.
Switching also takes time, albeit a very small time, but when the time
is shared across 32 clients it becomes very important.  Number of
devices, or download speeds, will be negatively affected by use of
Bluetooth.  Smartphones should be used with WiFi instead.

Did you test with Bluetooth turned off on the NUC?  Bluetooth consumes
shared resources; time, RF spectrum, antenna, radio, module memory,
and bus.

> PS the above unit comes with an "Intel® Dual Band Wireless-AC 3165"
> which we'll throw out as nec!  Or perhaps it suffices among the
> smallest installations, with only 12 simultaneous WiFi connections?

You can't tell.  It might work better with that motherboard.

> Thanks to anybody who can contribute to this R&D
> discussion/evaluation, as Intel NUC's soldered-down internal WiFi
> (likewise limited to 12 WiFi connections) has in the end become too
> constricting~

This isn't an R&D discussion; since your first post on 26th August,
I've seen no data from engineers as to what the underlying problem is.
No wireless packet traces, no kernel logs, no error logs, nothing.

I agree you are doing an evaluation.  On that basis, I agree with
Anish that a wireless router will do much better.  If equipment loss
is a problem, glue it to the inside of the case.  If the case isn't
big enough, bolt it to the outside of the case, or get a bigger case.

p.s. the Rpi3 wireless is working great for me too.

-- 
James Cameron
http://quozl.netrek.org/