[Rocks-Discuss] insert-ethers not working with power unit (PDU)

5 views
Skip to first unread message

Steven Berler

unread,
Jun 26, 2008, 8:27:10 PM6/26/08
to Discussion of Rocks Clusters
We just got an APC switched PDU that we want to add to our cluster to be
able to remotely turn off machines that are not in use. When I use
insert-ethers and select the "power units" appliance option, it never finds
the power switch. However, when I later tried to install compute nodes by
selecting the Compute option in insert ethers, it suddenly found the PDU and
identified it as a compute node (which assigned the PDU an ip address etc
and called it compute-0-0). How can I make the PDU be identified correctly
in insert-ethers? Ideally I would want it named something like power-0-0
and have it not identified as a compute node. Do I need to manually change
some of the xml files for setting up a custom appliance? Note the PDU is
indeed capable of sending BOOTP and DHCP requests. The only documentation I
could find on setting up a custom appliance concerned having a custom type
of compute node, not a PDU or other device. Obviously I don't need to make
a kickstart or anything for a PDU, all I need is to give it an ip and add
correct entrys to the Rocks database.

I was able to manually set the PDU ip by adding an entry to arp and using
telnet to log into the PDU. So possibly a fix would be to manually add
entrys to the Rocks MySQL database, but I had trouble finding what exactly I
would need to put in the database, and how to do this.

-Steven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080626/decca270/attachment.html

Greg Bruno

unread,
Jun 27, 2008, 11:19:42 AM6/27/08
to Discussion of Rocks Clusters
On Thu, Jun 26, 2008 at 5:27 PM, Steven Berler <ber...@gmail.com> wrote:
> We just got an APC switched PDU that we want to add to our cluster to be
> able to remotely turn off machines that are not in use. When I use
> insert-ethers and select the "power units" appliance option, it never finds
> the power switch. However, when I later tried to install compute nodes by
> selecting the Compute option in insert ethers, it suddenly found the PDU and
> identified it as a compute node (which assigned the PDU an ip address etc
> and called it compute-0-0). How can I make the PDU be identified correctly
> in insert-ethers? Ideally I would want it named something like power-0-0
> and have it not identified as a compute node. Do I need to manually change
> some of the xml files for setting up a custom appliance? Note the PDU is
> indeed capable of sending BOOTP and DHCP requests. The only documentation I
> could find on setting up a custom appliance concerned having a custom type
> of compute node, not a PDU or other device. Obviously I don't need to make
> a kickstart or anything for a PDU, all I need is to give it an ip and add
> correct entrys to the Rocks database.
>
> I was able to manually set the PDU ip by adding an entry to arp and using
> telnet to log into the PDU. So possibly a fix would be to manually add
> entrys to the Rocks MySQL database, but I had trouble finding what exactly I
> would need to put in the database, and how to do this.

my recommendation would be to power down all your compute nodes, then
remove all the compute nodes from the database. assuming you are
running rocks 5.0, for each compute node, execute:

# rocks remove host compute-0-0
# rocks remove host compute-0-1

etc.

after you remove the last compute node, execute:

# rocks sync config

then, start insert-ethers and select the 'Power Units' appliance.

then, power cycle your PDU (assuming that your frontend is not plugged
into the PDU). otherwise, telnet into your PDU (you said it got the
name compute-0-0, so you will need to execute 'telnet 10.255.255.254')
and run the command to reset the PDU.

then wait for PDU to be discovered by insert-ethers.

after it is discovered, stop insert-ethers, then restart it, but this
time select 'Compute' as the appliance.

- gb

Steven Berler

unread,
Jun 27, 2008, 3:31:03 PM6/27/08
to Discussion of Rocks Clusters
I was finally able to get it to come up in insert-ethers correctly. The
problem seems that I was not waiting long enough. I waited about 10 minutes
and finally it appeared in insert-ethers. During this entire time the
status light on the PDU was flashing orange, which according to the manual
means that it was continuously sending BOOTP requests.

The only question I still have, is do I need to wait for the isnert-ethers
list to display the (*) on the PDU or can I just force quit it?

-Steven

-------------- next part --------------
An HTML attachment was scrubbed...

URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080627/8c7b8d00/attachment.html

Anoop Rajendra

unread,
Jun 27, 2008, 3:38:48 PM6/27/08
to Discussion of Rocks Clusters
The PDU will most likely not require a kickstart file. So you may
safely force insert-ethers to quit.

-a

Tim Carlson

unread,
Jun 27, 2008, 4:47:53 PM6/27/08
to Discussion of Rocks Clusters
On Fri, 27 Jun 2008, Steven Berler wrote:

Just a note on these APC PDU's. They have a variety of options for setting
the DHCP address. You really need to get into the OS of the beast and tell
it to just do DHCP. Also if you have more than a few of these and plan to
do any SNMP queries, you are going to want to make sure they are all
running the same firmware. We have identical units that report quite
different SNMP information based on the firmware rev running on the PDU.

Good luck!!

-------------------------------------------
Tim Carlson, PhD
Senior Research Scientist
Environmental Molecular Sciences Laboratory

Steven Berler

unread,
Jun 27, 2008, 8:50:42 PM6/27/08
to Discussion of Rocks Clusters
Yeah, I was looking into the SNMP stuff. Is there any documentation on what
specific variables I need to be sending it in order to control the PDU? So
far all I've been doing is using snmpget and snmpwalk to just try to see
what is on there, but it seems like when I do snmpwalk I get a ton of copies
of the same data (ie I get like 10 different lists "Outlet 1" "Outlet 2"
..... etc). I've downloaded the PowerNet-MIB file, but I'm still new to
this SNMP stuff.

-Steven

URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080627/457de4f4/attachment.html

Tim Carlson

unread,
Jun 27, 2008, 9:39:50 PM6/27/08
to Discussion of Rocks Clusters
On Fri, 27 Jun 2008, Steven Berler wrote:

What APC unit do you have? The ones that we have are basic PDUs and you
can't control the individual outlets on the PDU.

I get power usage information by doing something like this

snmpget -v1 -c public power-0-2 -m "powernet388.mib" PowerNet-MIB::rPDULoadStatusLoad.1
snmpget -v1 -c public power-0-2 -m "powernet388.mib" PowerNet-MIB::rPDULoadStatusLoad.2
snmpget -v1 -c public power-0-2 -m "powernet388.mib" PowerNet-MIB::rPDULoadStatusLoad.3

Where depending on your version of firmware, 1 is the total, 2 is the
top bank, and 3 bottom bank. For example

snmpget -v1 -c public power-0-2 -m powernet388.mib PowerNet-MIB::rPDULoadStatusLoad.1
PowerNet-MIB::rPDULoadStatusLoad.1 = Gauge32: 138

Now you have to divide the answer by 10 to get the number of AMPs being
used.

Tom Rockwell

unread,
Jun 27, 2008, 10:07:46 PM6/27/08
to Discussion of Rocks Clusters
Hi,

Is there some description of what ROCKS can do with the APC stuff? We
have some networked APC fan doors and a stack of networked power
strips. When we used xCAT, it provided some expect scripts for outlet
control, but I haven't figured out what features ROCKS has for these.
(Actually I was surprised at the original message in this thread, I
never even thought to do an insert ethers on the APC stuff.)

Thanks,
Tom Rockwell
Michigan State University

Steven Berler

unread,
Jun 28, 2008, 3:06:49 AM6/28/08
to Discussion of Rocks Clusters
The APC unit we have is a "Switched PDU" model number AP7930. It has
individual control of each outlet, which so far I've only been able to
control using its telnet or web interface. It also has some nice features,
like having specified delays before changing outlets on/off. What we're
trying to do is eventually have the cluster automatically power down compute
nodes if theres no jobs in the queue. I'll probably end up writing some
scripts that extend the python scripts from the rocks-solid project, which
was made to work with IPMI.

URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080628/64af084c/attachment.html

Reply all
Reply to author
Forward
0 new messages