Skip to content

ISS-AF howto and usage documentation#

ISS Analisys Facility is a (small) cluster dedicated to local research activities (no GRID)

Cluster information#

Compute resources#

Particular nodes specifications: 

HOSTNAMES    PARTITION    S:C:T CPUS   MEMORY WEIGHT
issaf-0-0    CLUSTER     2:32:2  128   515000    100
issaf-0-1    CLUSTER     2:32:2  128   515000    100
issaf-0-2    CLUSTER     1:64:2  128   515000     50
issaf-0-3    CLUSTER     1:64:2  128   515000     50
issaf-0-4    CLUSTER     1:64:2  128   515000     50
issaf-0-5    CLUSTER     1:64:2  128   515000     50
issaf-0-6    CLUSTER    2:128:2  512  2097152     10
issaf-0-7    CLUSTER    2:144:2  576  2310755     10

Total resources: 

     Partition     #Nodes     #CPU_cores     Job_Nodes MaxJobTime Cores Mem/Node
     Name State     Total          Total     Min   Max  Day-hr:mn /node     (GB)
CLUSTER:*    up         8           1856       1          2-00:00   128     515+
      MPI    up         8           1856       1 infin    1-12:00   128     515+
      GPU    up         1            512       1 infin    1-12:00   512    2097

The issaf-0-6 node have two H200 NVidia GPUs and is has associated the GPU queue: 

HOSTNAMES    PARTITION    S:C:T CPUS   MEMORY WEIGHT
issaf-0-6    GPU        2:128:2  512  2097152     10


+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.71.05              Driver Version: 595.71.05      CUDA Version: 13.2     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA H200 NVL                Off |   00000000:D1:00.0 Off |                    0 |
| N/A   31C    P0             73W /  600W |       0MiB / 143771MiB |      0%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA H200 NVL                Off |   00000000:F1:00.0 Off |                    0 |
| N/A   32C    P0             74W /  600W |       0MiB / 143771MiB |      0%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+

        GPU0    GPU1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    224-231,480-487 28              N/A
GPU1    NV18     X      128-135,384-391 16              N/A

Storage resources#

Filesystem               Size  Mounted on
/dev/md126p1             1.8T  /export      <-- /home, RAID1 NVME
/dev/md123               1.8T  /software    <-- RAID1 HDD
/dev/md120p1              66T  /data        <-- RAID6 8xHDD
/dev/md121p1             200T  /data2       <-- RAID6 12xHDD
/dev/md119p1             200T  /data3       <-- RAID6 12xHDD
/dev/md124p1             200T  /data4       <-- RAID6 12xHDD
10.99.99.200:/storage01  200T  /data5       <-- RAID6 12xHDD
10.99.99.200:/storage02  200T  /data6       <-- RAID6 12xHDD
10.99.99.200:/storage03  200T  /data7       <-- RAID6 12xHDD

Resource usage#

Compute resources#

Warning

No compute loads, or long running processing should be run on the front-end node.
The front-end is both login and service node and this can impact overall usage.

Slurm scheduling is for the moment of type "first come, first served".
To be converted to fair share.

Storage resources#

  • /home should be used for small personal files, small analysis codes
  • /software is dedicated for software frameworks (locally compiled)
  • each user have an symbolic link $HOME/data pointing to /data/data_${USER} pointing to a much larger storage area

  • for requirements over 1TB, a request should be sent to administrator for creation of a personal directory in /data{2..7}

    Institute management approval required

    For data usage for more than 1 TB a request to institute management is required. First step is to send an email to institute director with cc to issaf AT spacescience DOT ro then more information will follow in subsequent e-mails.