On Sun, 27 May 2007 16:21:03 +0200
"Michal 'vorner' Vaner" <[EMAIL PROTECTED]> wrote:
> Hello
>
> AFAIK, I think you do not want to do that.
Not on important boxes, that's for sure!
> First: You need to load the kernel from the swap, in the time it
> loads, you have no running kernel (well, there is a little part, but
> that one has no clue about network).
My observations lead me to believe that, after going into hibernation
with swap suspend, the kernel boots up normally, and it's only at the
point of init running that it really comes 'out' of hibernation. As
far as running kernel before networking, it's a diskless host so the
kernel does do dhcp configuration at boot time.
> Second: You do not want swap on nfs, since it is terribly slow, buggy,
> nfs can allocate memory to transfer data and you get a circular
> problem
> - to get a memory, you need to get a memory. And, what if your cat
> steps on the ethernet cable?
Better risk that then constantly be running out of memory. I haven't
had a single problem with this. To be fair, it has 1/2 gig of RAM and
only 64M of swap, and although it only uses about a meg, it does use
that up pretty much right away.
> Third: the suspend does not use swap as a swap, but as a part of a
> disc.
As a block device you might say. A loop device is a block device like
any other, is it not?
> You might try suspend to ram, thought. It should work on diskless
> machine as well as on any other.
That is my preferred solution. Perhaps someday I will get it to work...
Anyhow, Hans-Werner-Hilse already provided a solution, which by the way
I haven't actually implemented, and probably never will, because it's
clearly a convoluted scheme in the first place. He suggested using an
initrd suspension, since you can get an initrd from a tftp server just
like a kernel image.
Thanks for the feedback,
Dan
--
[EMAIL PROTECTED] mailing list
