Syslinux - Nov 2005 - Consistent but intermittent DHCP problems after pxelinux loads kernel.

Question, testing the waters to see if anyone has input on this.  My 
summary belief is that I'm seeing a problem with the PXE bios code 
causing this issue.  I'm trying to find out what else I can do to 
confirm that, and rule out pxelinux/.  I'd like to get a confirmation 
that the use of pxelinux isn't a factor, as I'm not using
'keeppxe'.

I'm rolling out a couple hundred workstations, and using a pxelinux 
menu-driven update of the SBIOS and to do the OS install (SuSE 9.3). 
Hardware is HP xw9300 workstations,dual opteron, NVidia chipset and NIC. 
    The entire pxelinux-involved process works properly 100% of the 
time, DHCP, kernel loading, memdisk... etc.

The systems are showing one very improper behaviour:

Some significant portion of the time the DHCP OFFER is never REQUEST'ed, 
but DISC's and OFFERs keep flowing until it times out.  I've got packet 
captures of both successful and unsuccessful runs, with no difference in 
the decode results.  On failure, the client keeps requesting and 
ignoring the immediate offer that should be accepted is blindly ignored, 
and the 4- 8- 16- second retry is stepped through while the OFFER 
immediately comes through.

Looking at packet caps via a hub shared with the client, compared with 
the server logs shows nothing amiss.  This same exact image is 
successfully used to network install hardware from another vendor, as 
well as from a boot CD this same HP hardware.  The problem seems to 
recurr randomly, sometimes lasting a minute or two, sometimes hours.

Has anyone seen this or a similar issue on any configuration?  Any 
suggestions of something further I could try?  Thanks for any input.

Cheers,

/eli

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Fri, Nov 18, 2005 at 01:33:31PM -0800, Eli Stair
wrote:
> 
   <snip/> 
> The systems are showing one very improper behaviour:
> 
> Some significant portion of the time the DHCP OFFER is never
REQUEST'ed,
> but DISC's and OFFERs keep flowing until it times out.  I've got
packet
> captures of both successful and unsuccessful runs, with no difference in 
> the decode results.  On failure, the client keeps requesting and 
> ignoring the immediate offer that should be accepted is blindly ignored, 
> and the 4- 8- 16- second retry is stepped through while the OFFER 
> immediately comes through.
> 
> Looking at packet caps via a hub shared with the client, compared with 
> the server logs shows nothing amiss.  This same exact image is 
> successfully used to network install hardware from another vendor, as 
> well as from a boot CD this same HP hardware.  The problem seems to 
> recurr randomly, sometimes lasting a minute or two, sometimes hours.
> 
> Has anyone seen this or a similar issue on any configuration?  Any 
> suggestions of something further I could try?  Thanks for any input.
To me it seems the receive part does work.
 (possible causes: driver isn't aware of register change,
  interrupt isn't actived or misconfigured )


IIUC, it is the booting kernel DHCP code that fails.

To avoid DHCP in the early phase of start-up,
try IPAPPEND ( http://syslinux.zytor.com/faq.php )


GSt

P.S.

If it works, then contact me off-list for shipping an optron to me.
(I assume(hope?) that hpa has allready been rewarded by those
who use to software he wrote)

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)

iD8DBQFDfyVtOSINbgwa/7sRAj1XAJ9vZJLViaTcuEZpFXdS8PN3uNpTQgCfXRN+
Yzw0vHjUClxwwFgZSKTtlQk=b40j
-----END PGP SIGNATURE-----