Bryan J Smith wrote:
I'm currently starting to tackle this on two (2) fronts for RHEL:
1. Run Local (X+GNOME): PXE Stateless
2. Run Remote (X-only): LiveCD Tools (not for CDs, but EEPROM)
The distinction between the two really should be not much. Getting
/etc/rwtab* based readonly-root working properly on a distro should make
it work equally well with read-only Flash based boot or PXE network
read-only root boot.
I'm not against a hybrid mode either, where some things run local. But
right now I'm trying to find a balance, and possibly a competition,
performance-wise.
What do you mean by hybrid?
First off, it's clear there are initrd and init-script changes required
for RHEL in both solutions. E.g.,
https://fedorahosted.org/k12linux/wiki/RHEL5Server
I did extensive work trying to do this. While it is possible to
backport necessary pieces of mkinitrd to RHEL5, it was deemed infeasible
because it would forever be a fork. The changes necessary are too risky
and invasive to ever ship as a RHEL-5.x update.
I have some experience hacking on mkinitrd, and I think it would take me
~10 hours to backport and fully test. But even then I would be
uncomfortable with it because of the "forever a fork" problem.
As far as init-script changes, it would be rather minimal. I am under
the impression that mainly /etc/rwtab* tweaking can achieve most of what
is needed.
Just getting /tmp to come up as tmpfs seems to be a chicken-egg order
issue in PXE Stateless (using s-c-netboot, plus other mods).
I haven't seen this as a problem. How much have you tweaked /etc/rwtab*?
When it comes to the LiveCD Tools, there are things Anaconda doesn't do
in RHEL 5 that newer Fedora releases do. I don't think attempting to
bring in newer, select Anaconda packages (which a CentOS/Scientific
Linux effort is doing for 5) is the right move. I'd rather put it in a
%post or post-install modification from the stock/errata packages of
RHEL 5, to minimize engineering support requirements.
I did run into problems using livecd tools on RHEL5 due to missing
pieces of anaconda. In the end I gave up due to a combination of this
problem and the aforementioned backport issues.
I just got time at my client to start tackling these details head-on.
This is in addition to merging in select LTSP capabilities (LDM, LTSPfs
which requires FUSE) into RHEL 5, which I'm also tackling. And I won't
even go into the POS device details (that is another issue, one I am
tackling via another option for remote, they work fine locally in Linux,
which is what I spent some time playing with prior here at the client).
https://fedorahosted.org/k12linux/wiki/RHEL5Server
I built packages for FUSE and all this for RHEL5 here. I suspect the
FUSE part would just work if the user were added to the fuse group or
fusermount were setuid root. RHEL5 needs this while Fedora 9+ does not,
due to the new gvfs.
Warren Togami
[EMAIL PROTECTED]
_______________________________________________
Stateless-list mailing list
[email protected]
http://www.redhat.com/mailman/listinfo/stateless-list
