CentOS - Aug 2017 - Installation on Knights Landing (KNL) machines failure

Hi all,

We have a customer, where we are trying to install CentOS 7.3 on the new 
Intel Knights Landing nodes, unfortunately, we are unable to install 
then using PXE boot.

Now, I have a lot of experience of installation of machines using PXE 
boot, so the setup of that is not the problem. Both myself and my 
colleagues have looked through the problem, and we are unable to get 
much diagnosis of the problem.

When booting the machine via the network, it loads the ramdisk, and then 
we get no output, and then the machine reboots. Unfortunately, we have 
unable to get anything else on the screen. The machines don't have VGA, 
and monitor capable ports, so we are stuck with Serial Over Lan over IPMI.

We have however, after some searching via search engines, we found some 
documentation suggesting that RHEL 7.2 and then with significant updates 
on 7.3 do work on these types of machines. We have then followed the 
same process to try installing the machines, and we have managed to get 
the machine up, and in a working order. However, we'd like to use CentOS 
7.3, and would like to find the process of doing so, if it is any different

I was wondering to see if anyone else has seen any similar issues, and 
if there is a problem here, that we haven't tackled? Any further 
debugging we can do? Any assistance on this would be much appreciated

-- 
regards,

Arif Ali

Apologies for the noise, it was a mistake from our side

For some reason someone had set the installation directory to be from 
aarch64 instead of x86_64


On 22/08/2017 17:54, Arif Ali wrote:
> Hi all,
>
> We have a customer, where we are trying to install CentOS 7.3 on the 
> new Intel Knights Landing nodes, unfortunately, we are unable to 
> install then using PXE boot.
>
> Now, I have a lot of experience of installation of machines using PXE 
> boot, so the setup of that is not the problem. Both myself and my 
> colleagues have looked through the problem, and we are unable to get 
> much diagnosis of the problem.
>
> When booting the machine via the network, it loads the ramdisk, and 
> then we get no output, and then the machine reboots. Unfortunately, we 
> have unable to get anything else on the screen. The machines don't 
> have VGA, and monitor capable ports, so we are stuck with Serial Over 
> Lan over IPMI.
>
> We have however, after some searching via search engines, we found 
> some documentation suggesting that RHEL 7.2 and then with significant 
> updates on 7.3 do work on these types of machines. We have then 
> followed the same process to try installing the machines, and we have 
> managed to get the machine up, and in a working order. However, we'd 
> like to use CentOS 7.3, and would like to find the process of doing 
> so, if it is any different
>
> I was wondering to see if anyone else has seen any similar issues, and 
> if there is a problem here, that we haven't tackled? Any further 
> debugging we can do? Any assistance on this would be much appreciated
>
-- 
regards,

Arif Ali

Arif Ali wrote:
> Hi all,
>
> We have a customer, where we are trying to install CentOS 7.3 on the new
> Intel Knights Landing nodes, unfortunately, we are unable to install
> then using PXE boot.
>
> Now, I have a lot of experience of installation of machines using PXE
> boot, so the setup of that is not the problem. Both myself and my
> colleagues have looked through the problem, and we are unable to get
> much diagnosis of the problem.
<snip>
<RANT>
YES. I was unable to build my four nodes using PXEboot, because the damn
thing won't take what it's given, and won't skip to the default
target. It
INSISTS (ok, it is in the RFC, but...) on trying its MAC, or maybe it's
the UUID, I disremember, and spends *MINUTES - 4? 5? then tries again by
shortening it by one char, and again, and again, and by the time it tries
default, it's literally FIFTEEN MINUTES LATER, and the tftp/pxe has timed
out.
</RANT>

Had to use a USB.... Strong suggestion: if you're doing more than one
node, I've posted more than once how to do an upgrade by rsync. I did one
of our four, tried, and tried, and finally gave up, and once the first was
correctly built, rsynced to the other three nodes.

     mark

On Tue, 22 Aug 2017, m.roth at 5-cent.us wrote:
> <RANT>
> YES. I was unable to build my four nodes using PXEboot, because the damn
> thing won't take what it's given, and won't skip to the default
target. It
> INSISTS (ok, it is in the RFC, but...) on trying its MAC, or maybe it's
> the UUID, I disremember, and spends *MINUTES - 4? 5? then tries again by
> shortening it by one char, and again, and again, and by the time it tries
> default, it's literally FIFTEEN MINUTES LATER, and the tftp/pxe has
timed
> out.
> </RANT>
Mark,

I haven't faced the slow iterations you're seeing, but I sometimes use 
a shell script I wrote for naming PXE configuration files on per-IPv4 
bases:

   https://github.com/heinlein/pxehex

I don't know if it will help in your case, but I thought I'd pass it 
along.

-- 
Paul Heinlein
heinlein at madboa.com
45?38' N, 122?6' W