[TriLUG] booting off usbstick

[Joseph Mack NA3T via TriLUG]

On Sat, 11 Jul 2020, Stephen P. Schaefer via TriLUG wrote:

thanks for everyone's suggestions, but the task really is to boot an arbitary 
diskless computer off a usb stick, something that's done routinely with usb 
sticks with live file systems.

The problem is that I don't know how it's done. Just for fun, about 10 years 
ago, I tried booting my normal filesystem from a usb stick and got the same 
results (kernel panic) and gave up.

I'm back trying again. This time, the push is that I want the drive trays for 
storage, but this is incidental to getting the machine to boot off a usb stick.

> I may be doing something similar to what Joe is trying to do: I'm running 
> CentOS 8 from an (external) USB stick, with the four drive bays devoted to 
> providing NFS (stripe accros mirrors; (tested) spare drive in a desk drawer. 
> What's different is that I planned the system this way: I installed CentOS 
> from a DVD with none of the disk drives installed, just the USB, and CentOS 
> had no choice but to install on the USB.

I was in the process of building a usb stick from scratch this way, when I read 
your posting. I was heartened to know that in the right circumstances, it works. 
In my case, I didn't expect any joy, as this is the method I used to build the 
SATA disk that kernel panics booted off the usb bus. The install doesn't know 
whether it's writing to a usb stick or to a SATA drive, so I couldn't see how 
the files on the different types of devices would be any different. As it turns 
out, the usb stick I made this way kernel panics too.

The install makes an optional syslinux boot usb stick (which can boot your main 
install, in my case on a usb stick, if you choose not to install the standard 
boot loader on your installed disk). So I had two ways of booting the usb stick. 
The syslinux boot kernel panics too.

> It boots fine. When I added the disk drives later, they were recognized, and I 
> could configure them as I intended.

I'm happy for you.

> As to performance, once the OS gets loaded from USB, just about the only 
> access to the USB is to write log files.  There's plenty of RAM to cache the 
> OS file system, and to cache writes while they're waiting to get written.  In 
> my setup, the rate of writes to the USB is far less than its bandwidth, so I'm 
> having no performance trouble there.

I'm glad to hear performance isn't a problem. I don't care about performance. 
Current usb sticks are fast enough and the files are only going to be read once 
and sit in cache (32G RAM) after that.

This morning it occured to me that the difference in booting off a usb stick and 
a SATA drive is that I'm using the distribution's generic kernel, which likely 
won't have the usb drivers in it, but will load the usb drivers as modules 
later. (I used to make custom kernels with just the right drivers built into the 
kernel, but I gave up on this a while ago. Now I want installs that boot 
anywhere I put them.)

So I made the generic initrd (which I've never needed to use), using the 
supplied script, and this time the kernel stops with

can't mount /dev/sda1 on /mnt

and drops me into a shell. This is a step forward.

looking with ls, there are no /dev/sd*  in /dev

I'm now going to mess around around with initrd files and as well build a 
kernel with all the usb drivers built in.

If anyone knows the solution to the current situation I'd be glad to hear about 
it.

Thanks Joe

-- 
Joseph Mack NA3T EME(B,D), FM05lw North Carolina
jmack (at) wm7d (dot) net - azimuthal equidistant
map generator at http://www.wm7d.net/azproj.shtml
Homepage http://www.austintek.com/ It's GNU/Linux!