Hi Tim,
Of course, PXE booting is effectively DHCP controlled TFTP, so I'm
not sure you're out much with PVM here.
Since you can support any protocol of the host, I'm sure HTTP and
TFTP aren't much of a stretch. Luckily, just downloading and booting
the image off TFTP should get you going, if I understand correctly, then
that will using SMF I believe or perhaps a -B option from (py)GRUB, to
get the rest of the image from the HTTP based AI image server.
I'm admittedly ignorant of booting AI in HVM or PVM, Drew Fisher
and I were working on it a bit until he had to step out for a little
while as you know.
Thank you,
Clay
On Mon, 16 Feb 2009, Tim Foster wrote:
> Hi Sundar,
>
> On Fri, 2009-02-13 at 14:23 -0800, Sundar Yamunachari wrote:
>> pxe and DHCP is the only way to install OpenSolaris on X86 right now.
>
> Right,
>
>> What is the environment you have when you are booting PV guests.
>
> At the moment, jumpstart just uses NFS from the dom0 - PV installs get
> passed an nfs path to the jumpstart sysidcfg file, and the miniroot from
> the install image. We then use pygrub to boot this image.
>
> this example from virt-install(1):
>
> # virt-install \
> --paravirt \
> --name demo \
> --ram 500 \
> --file /export/guests/demo/images/demo.img \
> --file-size 6 \
> --nographics \
> --mac=00:16:3e:2f:8a:1a \
> --location
> nfs:netinstall:/export/s10u6/combined.s10x_u6wos/latest \
> --autocf=nfs:netinstall:/export/guests/demo/autocf
>
>
>> What protocols does it support?
>
> Well, this is all from the Solaris host, (the dom0) so theoretically,
> any protocol, just that the guest needs to get bootstrapped manually
> from what I can tell.
>
>> DHCP controls what boot programs the clients
>> will get along with IP address. You can tweak DHCP entries and use a
>> different boot file that is suitable for PV client. Further menu.lst
>> provides the information about the kernel and install programs needed to
>> perform the install. You need to have some mechanism to get that
>> information.
>
> Is menu.lst the only way to obtain this information?
>
>
> My problems booting a HVM guest continue:
>
>>> [1] the guest then fails in the network/physical start method I think,
>>> where an interface isn't plumbed automatically, dropping me to
>>> single-user maintenance mode. This is weird, since I'm able to manually
>>> plumb the xnf0 interface, and obtain a dhcp address. More investigation
>>> needed my end, I admit.
>
> I still can't work out what's going on here (and diagnosing a half-built
> miniroot is really painful!)
>
> - it looks like smf_netstrategy() in /lib/svc/share/smf_include is
> highly suspicious: it seems to only set $_INIT_NET_IF in the case where
> the root filesystem obtained via /sbin/netstrategy is "nfs" or
> "cachefs". In AI, we have a "ufs" root filesystem, so never set this
> evironment variable.
>
> That said, I can't work out how non-xen HVM installs are able to plumb
> their primary interface. Have any of you guys tried out HVM guests with
> AI yet?
>
> cheers,
> tim
>
> _______________________________________________
> caiman-discuss mailing list
> caiman-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
>
