Slurm manages some very heterogeneous clusters today (32/64 bits and
little/big endian in the same cluster), so this _should_ work. If you
increase the debug level on your daemons and commands that should help
identify what RPC has the problem.
Quoting Uwe Sauter <[email protected]>:
Then the only thing that comes to mind is a possible discrepancy
between 32bit and 64bit (presuming that your ARMS are 32bit).
Some (network) protocol structures might use field sizes that differ
between the architectures.
Regards,
Uwe
Am 18.02.2015 um 15:21 schrieb Luis Kornblueh:
Hi Uwe,
The ARMs are raspberrypis running raspbian and this is LittleEndian.
Cheerio,
Luis
On 2/18/15 2:20 PM, Uwe Sauter wrote:
Hi,
as far as I know, ARM could be either little endian or mixed endian,
depending on how old the design is…
From Wikipedia:
"SPARC historically used big-endian until version 9, which is bi-endian,
similarly the ARM architecture was little-endian before version 3 when
it became bi-endian, and the PowerPC and Power Architecture descendants
of POWER are also bi-endian."
So it might depends on the firmware or kernel if your ARM is working in
little endianess mode.
Regards,
Uwe
Am 18.02.2015 um 13:44 schrieb Luis Kornblueh:
Dear Aaaron,
thanks a lot, I can pick up some logging on Friday morning. The stuff is
in school.
Cheerio,
Luis
On 2/18/15 1:28 PM, Aaron Knister wrote:
Hi Luis,
That's surprising to me. Can you send any errors you got when the
x86_64 controller was talking to the ARM compute nodes? I'm wondering
if there's a field in one of the SLURN binary protocol structures
whose length is architecture dependent. From what I gather ARM is
little endian so I don't *think* it's an endianness issue.
Thanks,
Aaron
Sent from my iPhone
On Feb 18, 2015, at 3:16 AM, Luis Kornblueh
<[email protected]> wrote:
Hello,
I try to use slurm on a cluster buid by school kids. The controller
daemon (slurmctld) runs on a x86_64 box with Debian wheezy and the
associated Debian package. The compute nodes are ARM based running
as well Debian wheezy (slurmd). It seems that the slurmctld and
slurmd cannot communicate which other. If I move the controller to an
arm
platform everything works fine. For cross testing I added a x86_64
box as compute node and that one could communicate with the x86_64
base controller.
This allows to assume, I think, that there is some trouble with the
communication protocol as it is not architecture/hardware independent.
Could you give me any help on how to deal with this?
Thanks a lot,
Luis
--
\\\\\\
(-0^0-)
--------------------------oOO--(_)--OOo-----------------------------
Luis Kornblueh Tel. : +49-40-41173289
Max-Planck-Institute for Meteorology Fax. : +49-40-41173298
Bundesstr. 53
D-20146 Hamburg Email: [email protected]
Federal Republic of Germany
--
Morris "Moe" Jette
CTO, SchedMD LLC
Commercial Slurm Development and Support
