Re: [JLSE-Admins] Xeon Phi cards
OK Ben, Thanks a lot. --Judi On 03/12/2015 05:18 PM, Allen, Benjamin S. wrote:
Dancelot and Prancelot have been provisioned with RHEL7.0. I'd suggest not doing a full yum upgrade as it doesn't look like Intel supports RHEL's 7.1 kernel yet.
IPMI access is available using hostnames dancelot.fto.alcf.anl.gov and prancelot.fto.alcf.anl.gov from ftlogin1/2. Username: osrspace, pass: zae8Shu3. Both machines are setup with serial console redirect which can be used via: ipmitool -H dancelot.fto.alcf.anl.gov -e^ -U osrspace sol activate, use "enter^." to escape the console.
I've disabled warewulf provisioning for the two nodes, so now they're booting from their local drives. I've added sudoers access as well.
Please document any BIOS changes you make. Please don't change or delete the wwadmin (UID 3) IPMI user. Please don't change any of the IPMI networking settings.
Each host has DNS and a registered IP address for its MIC:
FTM 1GbE interfaces on the host's br0 which MPSS will create:
dancelot-mic0.ftm.alcf.anl.gov: 10.17.6.15 prancelot-mic0.ftm.alcf.anl.gov: 10.17.6.16
IB via SCIF
dancelot-mic0.ftw.alcf.anl.gov: 172.17.6.15 dancelot-mic1.ftw.alcf.anl.gov: 172.17.6.16
Do *not* deviate from using these IPs.
Ben
On Mar 12, 2015, at 11:24 AM, Allen, Benjamin S. <[email protected]> wrote:
I could install RHEL7 on the servers and then you could install MPSS as Intel supports RHEL7. I just don't have the time to integrate MPSS into a RHEL7 warewulf image at the moment.
Ben
On Mar 12, 2015, at 11:22 AM, Judicael A. Zounmevo <[email protected]> wrote:
RHEL6 will do for the Xeon Phi card. However, I looked at the config of the nodes (hosts); and they are pretty good (far better than the Breadboard that I'm using for regular Xeon); so I was thinking that a recent kernel would be awesome for using the hosts as well to play with cgroups. We will wait until RHEL7 becomes usable :) Thanks, --Judi
On 03/12/2015 10:55 AM, Allen, Benjamin S. wrote:
Would RHEL6 be OK? I don't have a Xeon Phi image (i.e. MPSS) with RHEL7 yet.
IB access is fine.
Ben
On Mar 11, 2015, at 8:53 PM, Kazutomo Yoshii <[email protected]> wrote:
Hi Judi,
Testbed description: We need to often customize the kernel. Access to massive compute resources (e.g. Intel MIC) as well as RDMA-enabled network devices are required as well. Prancelot and Dancelot fit the hardware description. Be more specific! e.g., Prancelot and Dancelot with a Xeon Phi card and a local hard drive. We need RHEL7 or equivalent to be installed on the local hard drive initially and we will maintain the local hard drive contents. IPMI is required to cycle the machine power. We have no plan for PXE booting these nodes.
Do you really need RDMA-enabled NICs at this point?
NOTE: Prancelot and Dancelot are currently a diskless system, which means you cannot modify default OS image.
- kaz
On 03/11/2015 11:33 AM, Judicael A. Zounmevo wrote:
Ok Kaz, thanks. How is the attached description in your opinion? Thanks, --Judi
On 03/11/2015 10:54 AM, Kazutomo Yoshii wrote:
Hi Judi,
Can you write a brief project description for a dedicated OS research, specifying Dancelot and Prancelot?
Here is a list of the current JLSE projects. http://jlse.anl.gov/projects/
- kaz
On 03/11/2015 10:50 AM, Judicael A. Zounmevo wrote:
Hello Kaz, Yep it does. So are the nodes going to b flagged "reserved for OS research" after all? Thanks, --Judi
On 03/11/2015 10:33 AM, Kazutomo Yoshii wrote:
Hi Ben,
Thank you for installing the cards,
I think the JLSE provisioning system is not available for us, so we probably need a local hard drive plus IPMI.
Swann and Judi: does it sound good for you guys?
- kaz
On 03/10/2015 06:05 PM, Allen, Benjamin S. wrote:
These have been installed in Dancelot and Prancelot. Let me know if privileged access is needed on the machines, and if you'd like me to announce that these machines are dedicated for OS Research use. Currently the two machines are diskless provisioning from JLSE's Warewulf. I could make them statefully installed if you want. Also note that these machines are on the normal JLSE "FTM" and "FTW" networks, so please no DHCP servers, etc.
Ben
On Mar 9, 2015, at 9:37 AM, Kazutomo Yoshii <[email protected]> wrote:
Swann,
I'm talking about other Phi cards for JLSE nodes, not knight. It will not affect your work.
- kaz
On 03/09/2015 09:18 AM, Swann Perarnau wrote:
Hi,
I'm still running some tests on it today, can we wait until tomorrow to replace the cards ?
Thanks,
Swann
On Sat, Mar 07, 2015 at 06:18:03PM -0600, Kazutomo Yoshii wrote:
Hi Ben,
I left two 16GB Xeon Phi cards on your workspace in tinkerlab.
Please replace the old Phi cards with the new ones, so that Swann and Judi can use the new cards. I may also occasionally use the Xeon Phi.
Here are the serial numbers (probably not needed but just in case). ADKC32200717 ADKC31304009
- kaz
Admins mailing list [email protected] https://lists.jlse.anl.gov/mailman/listinfo/admins
Admins mailing list [email protected] https://lists.jlse.anl.gov/mailman/listinfo/admins
participants (1)
-
Judicael A. Zounmevo