🚀 Onboarding - IBM Fusion HCI into COE Lab 🥳

[rbo]

We're getting an IBM Fusion HCI loaner!

Rough hardware specs:

3x Fusion HCI Storage Server (9155-C10)
    - 2x 7.68TB NVMe PCIe5.0,
    - 2x 960GB M.2 NVMe OS-Laufwerke (RAID 1)
    - 32 Cores (2x 16-Core Intel Gold 6426Y 2.5GHz 185W )

3x Fusion HCI Storage Server (9155-C10)
    - 2x 7.68TB NVMe PCIe5.0
    - 1x 1.6TB Local Storage Drive
    - 2x 960GB M.2 NVMe OS-Laufwerke (RAID 1)
    - 256 GB RAM
    - 32 Cores (2x 16-Core Intel Gold 6426Y 2.5GHz 185W )

1x Fusion HCI Storage Server (9155-C14) - 
    - 2x 960GB M.2 NVMe OS-Laufwerke (RAID 1)
    - 1024 GB RAM
    - 64 Cores (2x 32-Kern Intel Gold 6438 2.0GHz 205W Prozessoren)

1x Fusion HCI Service Node 

1x IBM SF GPU Server (9155-G03) 
    - 2x M.2 960GB NVMe drives for the operating system and the drives are in a RAID 1 configuration
    - 2x AMD EPYC 9254 24C 200W 2.9GHz
    - 768GB RAM (24x 32GB TruDDR5 4800MHz (2Rx8) RDIMM-A)
    - 4x Nvidia L40S 48 GB GPU

We have to rip out other hardware and plan everything for the onboarding.

ToDo's

• Planning Power consumtion
• Planning delivery with GWS
• Planning rack space
  • ➡️ Right rack top 1/2 of the rack.
  • Relocate the servers
    • What is with switch dgs-3630-1?
    • inf3 -> @dmoessne move stuff to a VM
    • ucs56 -> Remove from ISAR cluster
    • inf1 -> Time for RHEL10 - core services (inf1/inf31) #288 - move to sf1
    • inf44 -> powered of - ready to store in the basement
    • inf31 -> Time for RHEL10 - core services (inf1/inf31) #288 - move to a VM
  • Or Maybe remove because of power consumtion

[rbo]

Hardware and AVX2 (RHEL10 readyness)

# ocp5
% oc get nodes -Lcpu-feature.node.kubevirt.io/avx -L cpu-feature.node.kubevirt.io/avx2
NAME                 STATUS                        ROLES                        AGE      VERSION   AVX    AVX2
ocp5-control-0       Ready                         control-plane,master         3y118d   v1.32.7   true   true
ocp5-control-1       Ready                         control-plane,master         3y118d   v1.32.7   true   true
ocp5-control-2       Ready                         control-plane,master         3y118d   v1.32.7   true   true
ucs-blade-server-5   Ready                         nodeAgent,worker             2y150d   v1.32.7   true   true
ucs-blade-server-6   Ready                         nodeAgent,worker             2y150d   v1.32.7   true   true
ucs-blade-server-7   NotReady,SchedulingDisabled   nodeAgent,worker             2y255d   v1.32.7   true   true
ucs-blade-server-8   Ready,SchedulingDisabled      nodeAgent,worker,worker-hp   2y255d   v1.32.7   true   true
win-nje44grs2gf      NotReady,SchedulingDisabled   worker                       2d21h    v1.32.6
# isar
% oc get nodes -Lcpu-feature.node.kubevirt.io/avx -L cpu-feature.node.kubevirt.io/avx2
NAME     STATUS   ROLES                  AGE    VERSION   AVX    AVX2
ceph10   Ready    storage-node,worker    79d    v1.32.7   true
ceph11   Ready    storage-node,worker    443d   v1.32.7   true
ceph12   Ready    storage-node,worker    443d   v1.32.7   true
inf4     Ready    control-plane,master   704d   v1.32.7
inf5     Ready    control-plane,master   704d   v1.32.7
inf6     Ready    control-plane,master   704d   v1.32.7
storm2   Ready    virt-node,worker       145d   v1.32.7   true   true
storm3   Ready    virt-node,worker       145d   v1.32.7   true   true
storm5   Ready    virt-node,worker       145d   v1.32.7   true   true
storm6   Ready    virt-node,worker       145d   v1.32.7   true   true
ucs55    Ready    virt-node,worker       38d    v1.32.7   true
ucs56    Ready    virt-node,worker       38d    v1.32.7   true
ucs57    Ready    virt-node,worker       38d    v1.32.7   true

[rbo]

Space is available. Power is provided by DIRECT A&B - We have to power of some ucs nodes and/or the entire ucs blade center.

[DanielFroehlich]

I vote to power off UCS nodes first - we can compensate that easily on ISAR by moving workload to the storm servers. The blades are used by OCP5 and VMWare and would probably be missed a lot.

[stefan-bergstein]

I vote to power off UCS nodes first - we can compensate that easily on ISAR by moving workload to the storm servers. The blades are used by OCP5 and VMWare and would probably be missed a lot.

Yes, a lot. Resources can be easily compensated... A learning playground with admin access not :-/

Anyway, I am open for alternatives. E.g. assisting as RHOAI admin on the HCI cluster.