[Rocks-Discuss] Installation problem with HP rx2660 integrity server

[Madhurjya P Bora]

Hi All!

I am trying to make a ROCKS cluster as a x86-64 machine as the frontend and
ia64 (HP Integrity rx2660 Server) machines as computing nodes. I have chosen
to install ROCKS 4.2.1 as it supports both x86-64 and ia64. The
frontend installed
smoothely and after I rebooted I also did a copyroll on the ia64 CDs
and made the
distribution with arch=ia64 (with rocks-dist).

I then put on the ia64 kernel-roll CD on one of the computing node and started
kickstarting. The insert-ethers correctly detected the node and
kickstarting started
successfully. However right after /sbin/install command on the computing node
the anaconda failed.

I then tried to install the compute node as a frontend with the ia64
kernel-roll CD
but same happened. The installation halted after anaconda started. But the CD
was properly copied and the downloaded ia64 kernel .iso was properly checked
with md5sum

What might be the reason, I am at my wits end. Is it a memory problem (i.e.
should I do a frontend mem=1024MB installation) or it is a CD burning problem?
However if I try to load a separate RHEL-5 (AS) (which I have) on the
compute node,
it does it correctly.

The ia64 computing nodes have single dual core Itanium 1.8 GHz CPUs with 4GB
of RAM each and 146 GB SAS HD with Raid-1. The frontend is a Intel Dual Core
machine (2.8 GHz) with 1GB of RAM. It has two NICs, one which was inserted
as a PCI card and is connected to the compute nodes via Gb switch and the other
which is integrated but does not have a driver (at this moment) for
the Centos-4 O/S
that came with ROCKS 4.2.1. Could this be a problem, as the other NIC which
is to be connected to the outside LAN is not detected but not
installed on the frontend.

Any suggestion will be highly appreciated. Regards,

-- Madhurjya
------------------------------------------------------
Dr Madhurjya P Bora
Physics Department, Gauhati University
Guwahati 781 014, India.
mpbora AT gauhati.ac.in
------------------------------------------------------

[Greg Bruno]

on the itanium nodes, try installing it with CentOS 4 update 4 (that
is the release that rocks 4.2.1 is based on). if the node installs,
then we know it isn't a driver issue.

- gb

[Madhurjya P Bora]

It seems that anaconda is not finding the display and qutting with an
error. But with
correct kickstarting as compute node, it should not complain for
display! Any suggestion?

--- Madhurjya

--

------------------------------------------------------
Dr Madhurjya P Bora
Physics Department, Gauhati University
Guwahati 781 014, India.

Phone : +91-361-2570531 (O)
+91-361-2574165 (R)
Fax : +91-361-2700311
------------------------------------------------------

[Roy Dragseth]

You might want to check the console setting on your node. On bootup look at

Boot Option Maintenance Menu
-> Select Active Console Output Devices

It should look something like this

EFI Boot Maintenance Manager ver 1.10 [14.61]

Select the Console Output Device(s)

Acpi(HWP0002,0)/Pci(1|1)/Uart(9600 N81)/VenMsg(PcAnsi)
Acpi(HWP0002,0)/Pci(1|1)/Uart(9600 N81)/VenMsg(Vt100)
* Acpi(HWP0002,0)/Pci(1|1)/Uart(9600 N81)/VenMsg(Vt100+)
Acpi(HWP0002,0)/Pci(1|1)/Uart(9600 N81)/VenMsg(VtUtf8)
* Acpi(HWP0002,0)/Pci(4|0)
Save Settings to NVRAM
Exit

I do not quite remember why this was needed, but it seems like anaconda bails
out if it only finds a serial console.

r.

[Madhurjya P Bora]

Finally, I have been able to install the computing nodes. I had to set
the primary
display to the VGA (though a monitor was not connected) and the secondary to
the serial console. I believe, the ROCKS developers should streamline this
in anaconda, so that it should not complain about not having a VGA display
whle installing a compute node.

One more querry! After installation, I started ganglia, but after a short while
it shows the compute nodes as down though they are up and I can ssh to them.
This problem repeatedly happens and I have not been able to troubleshoot it.
Any ideas?

--- Madhurjya

On Fri, Apr 4, 2008 at 10:54 PM, Madhurjya P Bora

[Philip Papadopoulos]

On Tue, Apr 8, 2008 at 8:13 AM, Madhurjya P Bora <madhurj...@gmail.com>
wrote:

> Finally, I have been able to install the computing nodes. I had to set

> the primary
> display to the VGA (though a monitor was not connected) and the secondary
> to
> the serial console. I believe, the ROCKS developers should streamline this
> in anaconda, so that it should not complain about not having a VGA display
> whle installing a compute node.

Headless nodes are supported in Rocks 4.3. But we no longer have any itanium
hardware so it is quite difficult for us to test itanium-specific issues.
Itanium EFI bios is quite different from usual PC bios.

>
>
> One more querry! After installation, I started ganglia, but after a short
> while
> it shows the compute nodes as down though they are up and I can ssh to
> them.
> This problem repeatedly happens and I have not been able to troubleshoot
> it.
> Any ideas?

Yes. Your switch is probably selectively blocking multicast traffic. Usually
through overly aggressive IGMP snooping.

-P

--
Philip Papadopoulos, PhD
University of California, San Diego
858-822-3628
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080408/55979a23/attachment.html

[Madhurjya P Bora]

Is there any solution for it? I mean does it mean that ganglia will never
show any true status of the machines? I am using an 8-port D-Link
Gigabit switch for the interconnect.

-- Madhurjya

>
>
>
> On Tue, Apr 8, 2008 at 8:13 AM, Madhurjya P Bora <madhurj...@gmail.com>
> wrote:
> > Finally, I have been able to install the computing nodes. I had to set
> > the primary
> > display to the VGA (though a monitor was not connected) and the secondary
> to
> > the serial console. I believe, the ROCKS developers should streamline this
> > in anaconda, so that it should not complain about not having a VGA display
> > whle installing a compute node.
> Headless nodes are supported in Rocks 4.3. But we no longer have any itanium
> hardware so it is quite difficult for us to test itanium-specific issues.
> Itanium EFI bios is quite different from usual PC bios.
>
>
>
>
> >
> >
> > One more querry! After installation, I started ganglia, but after a short
> while
> > it shows the compute nodes as down though they are up and I can ssh to
> them.
> > This problem repeatedly happens and I have not been able to troubleshoot
> it.
> > Any ideas?
> Yes. Your switch is probably selectively blocking multicast traffic. Usually
> through overly aggressive IGMP snooping.
>
> -P

--

[Mason J. Katz]

You'll need to see if you can configure your switch. It should have a
web interface where you can see the IGMP settings.

On Wed, Apr 9, 2008 at 9:05 AM, Madhurjya P Bora

--
-mjk