[Server-devel] [XSCE] Re: Root fs on XO1

[George Hunt]

On Sun, Aug 11, 2013 at 6:49 PM, James Cameron <quozl at laptop.org> wrote:

>
> Rather, I guess that our dracut-modules-olpc package simply hasn't got
> scripted support for swinging the kernel over to a root filesystem on
> the SD card on XO-1.
>
> http://dev.laptop.org/git/users/dsd/dracut-modules-olpc/
>
>
> http://dev.laptop.org/git/users/dsd/dracut-modules-olpc/tree/30olpc-boot/olpc-boot-cmdline.sh
>
> The above script is where the decision is made, and XO-1 SD card is
> not in the case statement, because we didn't add this support.
>
> So yes, specifying a root filesystem using the command line may be a
> workaround.
>

Really helpful information.


>
> > Then I started following James' suggestion, looking into
> > olpc-os-builder.   There's an sd_image option which can be
> > enabled.  I could not determine if enabling that would also enable
> > booting from an external USB flash/hard disk, but it seems worth a
> > try.
>
> The lack of support in dracut-modules-olpc would still need fixing, in
> my opinion.
>
> The sd_image option will make a .zd but XO-1 doesn't have fs-update.
>
> > (I'm still toying with the idea of a WD Passport -- external hard
> > disk as the supported option for the XO1 XSCE).
>
> I've avoided that brand because of extra features of the product, in
> case they interfere.  Does it work okay?
>

I was really fortunate, in my testing yesterday.  I happened to have a USB
ethernet dongle, and a 5v tplink 3020 wifi unit attached when I started
testing.  The WD disk did not spin up.  I removed the ethernet and wifi
loads, and it spun up correctly. The fortunate part was that a
vulnerability was exposed, and the opportunity to address it is at hand.

I'm going to spend some time today to get numbers for the interfering
loads. The XO4 under test is connected to a "watts up" power meter. This
morning, idle, the watts-up indicated 3.3w. I did an "ls" on the mounted
drive, and saw a transient on the meter to something between 7 and 8 watts.
 I'm sure I, and the meter, didn't see the whole event. But after a length
of time (probably spin up) the power was 4.6 (maybe the additional cpu and
spinning current). Letting the XO4, and the hard disk, go to idle, I
removed the hard disk USB connector.  After a short time, the power
decreased (after  a long time at idle) from 4.0w to 2.2w.

So the "watts-up" is probably not very accurate, but it yields this set of
observations:"

   - The spin up creates a transient on USB power of almost an amp.
   - The spun down power,on idle, drawn by the WD hd is about 2 watts
   (4.0-2.2watts; =>1.8w / 5V = .36A)
   - Failure to provide stiff enough power on USB 5v leads to HD
   initialization failures
   - Whether in the WD drive itself, or the OS, there will be spin down/up
   events during normal server operation
   - Adding back in the wifi adapter, and ethernet dongle raised the idle
   power to 5.6 Watts (from 4.0w).
   - The power to keep the disk spinning is apparently about .6w.
   - When I disconnected the tplink mr3020 there was a .6w change in power
   => .6w / 5V = .12A = 120mA (idle--no RF activity)


So I ordered 2 12v USB hubs from amazon to do some additional testing:
http://www.amazon.com/Anker%C2%AE-Uspeed-4-Port-Adapter-Chipset/dp/B005QWY3PU
http://www.amazon.com/StarTech-com-Mountable-Rugged-Industrial-ST4200USBM/dp/B001H7AFO6/

I had also ordered, earlier, the following items, which I think might be
needed in a deployment where the school server needs to be available during
the school day, even when the power happens to be off:
http://www.amazon.com/gp/product/B003QRQ5DK/
http://www.amazon.com/gp/product/B00A71CMDU
http://www.amazon.com/gp/product/B003UANXCE/

I'm beginning to solidify in my mind a system which includes some or all of
the above, which I can begin to test. I've added to the list a Sears deep
cycle battery, which I bought locally for about $130.

I had concluded, earlier, that the loop resistance of a 300 ft max cat5
ethernet hub-to-hub link is about 8 ohms. With the 120mA observation above,
for the power consumed by the tplink mr3020, we'd expect a .96V supply
droop running mr3020's at max distance -- not likely to work.  That's why I
think running power-over-the-internet at 12v with the 12v dc to 5
converter, seems attractive.  But it will take more experimentation with
mr3020's to see how much supply droop they can handle -- and where the
cable length begins to require dc conversion.

(Extra features of consumer backup drives _might_ involve changes to
> the device in terms of the USB descriptors, such as those required for
> a push button to start backup software, or CD-ROM images of the
> software, so I've tended to select drives that either don't mention
> the software, or are marketed as expansion rather than backup.)
>
> > I'm not yet clear whether it would be any simpler to insist that
> > XO1's be unl
>
ocked before installing XSCE. It's easy to modify the
> > kernel command line via "olpc.fth", (which I verified to be ignored
> > on a locked XO) but maybe getting the right drivers in the initramfs
> > is the hard part.  I still have lots more to learn.
>
> I really suggest you disable-security.  Unless you have a real or
> industrial need for the security system, it will just hurt you over
> and over.
>
> --
> James Cameron
> http://quozl.linux.org.au/
>

I've noticed that just hitting the projecting HD USB connector can cause a
disconnect/reconnect cycle displayed on an open console window.  Does this
often crash an external disk? Does this observation favor ext3 or ext4 over
ext2?

There's so much I don't know.
George
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.laptop.org/pipermail/server-devel/attachments/20130812/aa696dbd/attachment-0001.html>