Re: Re[2]: linux boot time

Re: Re[2]: linux boot time

[David Woodhouse]

rsmith@bitworks.com said:
>  I think I posted some info about my experiences a while back.... that
> may have been on the embedded list though...

> David, how many other reports besides mine to you have of the CF's
> developing a problem?  We use both SanDisk and SST CF's in a couple of
> projects and most of our failures are during the development cycle.
> Go figure... 

There was Vipin Malik's writeup at http://www.embeddedlinuxworks.com/ but 
that doesn't seem to be there any more. He was actually doing powerfail 
testing on JFFS/JFFS2 and making me fix it when it broke, but he also did 
some on CF, and obviously could do nothing when it broke. Software is _so_ 
much nicer.

Charles Manning's rants about bog-roll technology¹ are too vehement to have 
been born from anything but personal experience as well, AFAICT :)

--
dwmw2
¹ http://mhonarc.axis.se/jffs-dev/msg01267.html

Getting big flash onto motherboards. Will CF work? what will?

[Ronald G Minnich]

What's the right way to get a reasonably-sized flash part onto a commodity
motherboards, on a cluster of size 1024 nodes.

here's my problem. I'm setting up a linuxbios-based cluster of 1024 nodes.
I need about 2 Mbytes of flash, so I can get a kernel, initrd, and myrinet
driver in there for linuxbios to load and boot.

We don't want etherboot:  tried it, don't like it, won't work on myrinet
anyway. So we need a bigger flash part than is available on any P4
motherboard.

I can't just use DoC, as we have done on some motherboards, because it
takes way more than 512 or 1024 bytes of code to turn on those intel
chipsets. Also these boards use a Firmware Hub and the signalling is a bit
different (I think) than a standard FLASH part.

I think I'm stuck leaving the 512Kbyte flash part in there. Hence I need
to get an ADDITIONAL flash part on these motherboards. I had thought to
use IDE->CF adapters and put 1024 CF cards in the cluster. Note that these
parts would be read-only: on our Alpha cluster, we last updated flash 10
months ago. Some of the comments about CF reliability are a case for
concern.

Are there other options? These are 1U or .8U spaces most likely, and so a
big PCI card is out of the question -- that's needed for Myrinet. There
are no ISA slots. There are two IDE headers, and possibly a SCSI header. A
very low profile PCI card might work. The only option I was aware of was
IDE->CF adapaters. Is there something else I should look at?

thanks

ron

Re: Getting big flash onto motherboards. Will CF work? what will?

[David Woodhouse]

rminnich@lanl.gov said:
>  Are there other options? These are 1U or .8U spaces most likely, and
> so a big PCI card is out of the question -- that's needed for Myrinet.
> There are no ISA slots. There are two IDE headers, and possibly a SCSI
> header. A very low profile PCI card might work. The only option I was
> aware of was IDE->CF adapaters. Is there something else I should look
> at?

Given that you can ensure the BIOS won't get too upset -- could you glue 
some flash chips onto a DIMM?

If they're going to be read-only, I suspect you'd get away with CF though.

--
dwmw2

Re: Getting big flash onto motherboards. Will CF work? what will?

[Ronald G Minnich]

On Fri, 17 May 2002, David Woodhouse wrote:

> Given that you can ensure the BIOS won't get too upset -- could you glue
> some flash chips onto a DIMM?

We don't want to build this. Can we buy it?

> If they're going to be read-only, I suspect you'd get away with CF though.

Good to know, thanks.

ron

Re: Getting big flash onto motherboards. Will CF work? what will?

[David Woodhouse]

rminnich@lanl.gov said:
>  We don't want to build this. Can we buy it?

Dunno; I just made it up -- is it even technically possible to put flash in 
the DIMM slots of most motherboards?

--
dwmw2

Re: Getting big flash onto motherboards. Will CF work? what will?

[Ronald G Minnich]

On Fri, 17 May 2002, David Woodhouse wrote:

> Dunno; I just made it up -- is it even technically possible to put flash in
> the DIMM slots of most motherboards?

I think on SDRAM it would be doable, but i can't see anyone putting in the
effort as you have to get all the state machines right, and then you have
to convince the BIOS you're not really there.

In an earlier life I did design 2 different NICs for DRAM slots, and it
was not hard. But that was before SDRAM and its complexities.

ron

Re: Getting big flash onto motherboards. Will CF work? what will?

[Russ Dill]

> Are there other options? These are 1U or .8U spaces most likely, and so a
> big PCI card is out of the question -- that's needed for Myrinet. There
> are no ISA slots. There are two IDE headers, and possibly a SCSI header. A
> very low profile PCI card might work. The only option I was aware of was
> IDE->CF adapaters. Is there something else I should look at?

there are DOC devices that plug into IDE

Re: Getting big flash onto motherboards. Will CF work? what will?

[David Woodhouse]

Russ.Dill@asu.edu said:
>  there are DOC devices that plug into IDE

Those are DiskOnModule -- they're effectively CF too, although maybe in a
different form factor. Although M-Systems do seem to have some clue about
how to do such things so perhaps they've managed to produce a _reliable_ CF
device.

--
dwmw2

Re: Getting big flash onto motherboards. Will CF work? what will?

[Richard A. Smith]

On Fri, 17 May 2002 09:57:31 -0600 (MDT), Ronald G Minnich wrote:

>Are there other options? These are 1U or .8U spaces most likely, and so a
>big PCI card is out of the question -- that's needed for Myrinet. There
>are no ISA slots. There are two IDE headers, and possibly a SCSI header. A
>very low profile PCI card might work. The only option I was aware of was
>IDE->CF adapaters. Is there something else I should look at?

The IDE->CF is probally your best option. If you can't find a small
PCI card with flash on it.
I think that in read only mode you probally won't have any issues. 
Especially since these things won't power cycle much...

However if you do have  a problem...

<Shameless Plug>
Bitworks, Inc has an IDE to CF adapter that we have used in some of
our products.  We call it the BitDrive.

We could modify this board for you so that it ran off the 12V supply
rather than the 5V and add in a 5V Linear LDO regulator with a large
cap on the 5V rail such that the CF would have lots of clean power
long after the main power died.  This should insure that the CF's
don't die due to power issues.

In qty. of 1000 I'm pretty sure we can be pretty close to the cost of
purchasing an off-the-shelf adapter.

Give me me a ring if your interested.

BTW.. We could also build you a low profile Flash memory PCI card but
I think that will be more expensive.

</Shameless Plug>
--
Richard A. Smith                         Bitworks, Inc.               
rsmith@bitworks.com               479.846.5777 x104                        
Sr. Design Engineer        http://www.bitworks.com   

Re: Re[2]: linux boot time

[Richard A. Smith]

On Fri, 17 May 2002 16:44:43 +0100, David Woodhouse wrote:

>There was Vipin Malik's writeup at http://www.embeddedlinuxworks.com/ but 
>that doesn't seem to be there any more. He was actually doing powerfail 
>testing on JFFS/JFFS2 and making me fix it when it broke, but he also did 
>some on CF, and obviously could do nothing when it broke. Software is _so_ 
>much nicer.

Hmmmm... Do you have an e-mail address for him? I would like to get
info on the type of CF used and the specific failure mode.   

Was he testing JFFS on a CF or just powercycleing a CF under use?


--
Richard A. Smith                         Bitworks, Inc.               
rsmith@bitworks.com               501.846.5777 x204                        
Sr. Design Engineer        http://www.bitworks.com