Re: More on iSCSI boot

[julian_satran@il.ibm.com]

David,

The trouble with our boot is two-fold:

-the SCSI boot - unlike the network boot is unbounded (there is no such
thing as an image).
-even if we would like to standardize a "primary" or "minimal" boot we have
no good understanding (experience)
of how this will interact with the iSCSI security mechanisms.

Julo

Black_David@emc.com on 31-05-2001 05:49:57

Please respond to Black_David@emc.com

To:   ips@ece.cmu.edu
cc:
Subject:  More on iSCSI boot




Let's start with a simple boot from a disk - the system
BIOS reads the boot sector(s) off of the disk drive,
loads and runs it/them.  That primary bootloader then
handles whatever else is necessary based on the ability
to do disk reads (a secondary bootloader and/or other
things may be involved).  On Intel systems, it's generally
a combination of the system BIOS and card BIOS that
make the disk reads work.  Simplifying assumptions
are common (e.g., boot from LUN 0 on the first SCSI
target found).  The goal of the iSCSI boot draft is
to explain how to make this "simple boot from a disk"
mechanism work when the boot disk is attached via iSCSI.

An iSCSI adapter has some things to do in order to make
this work, e.g., it has to log into the target before disk
reads can be issued.  I tend to believe that "Don't do this
(i.e., try to boot over iSCSI)" is not an acceptable answer.

In the message that started this whole boot discussion, I
suggested that

     Using DHCP to find SLP to find the boot device seems
     both clumsy and an invitation to problems (one more
     thing that can break and prevent booting),

I think that's doubly true if LDAP is used instead of SLP.
David Robinson's messages support my inclination to reuse
DHCP option 17 (Root Path) by defining iSCSI syntax
for it.  Between that, any TCP parameters that one wants to
set through DHCP, and iSCSI parameters that can be defaulted,
it should be possible to get the first disk reads done
through iSCSI.

As has been pointed out, booting is often poorly secured in
general.  While it'd be nice to change this, iSCSI will face
the same pressures that other boot mechanisms face - keep
it simple, get the boot image loaded, and let it do something
fancier.  Any signature/validation of the boot image will be
above the level of iSCSI.  Somehow, I don't expect to find
implementations of ESP in card BIOSes anytime soon.

If implementers want to use DHCP and TFTP to boot, I don't
see any point in stopping them, but I don't think either should
be mandated.  DHCP has centralized administration advantages,
and TFTP is a simple way to download code, but both are "one
more thing that can break and prevent booting" and hence may
not be used all the time.

Comments?
--David

---------------------------------------------------
David L. Black, Senior Technologist
EMC Corporation, 42 South St., Hopkinton, MA  01748
+1 (508) 435-1000 x75140     FAX: +1 (508) 497-8500
black_david@emc.com       Mobile: +1 (978) 394-7754
---------------------------------------------------