Is there any benchmark running on HPC cluster?

General Discussion
1

There is a newly built HPC cluster that contains hundreds of Intel Xeno Goal 6338 and all connected with Infiniband. But I find that the the operating system install in this cluster is CentOS 7.4, which was released about 5 yrs ago. Its kernel version is still 3.10. I found in this specification said that the minimal kernel version should be 4.12. The argument of the service provider about why they chose CentOS 7.4 is to ensure the software compatibility, but I have concern that there is unnecessary performace losses due to using lagacy linux system on the latest hardware.

I find that there are serveral benchmarks (e.g.) running on single node which show significant performance improvement by using ClearLinux. I am wondering is there any benchmark that run on HPC cluster? BTW, is ClearLinux suitable for HPC scenario? It would be really appreciated to know some real use cases.

2

Hi @link89 ,

I am not sure if my two nodes classify as HPC, since I don’t run calculation accross the nodes. Anyway, the output from neofetch:

          BBB                rene@n001 
       BBBBBBBBB             --------- 
     BBBBBBBBBBBBBBB         OS: Clear Linux OS x86_64 
   BBBBBBBBBBBBBBBBBBBB      Host: ProLiant ML350 Gen10 
   BBBBBBBBBBB         BBB   Kernel: 5.10.30-1032.native 
  BBBBBBBBYYYYY              Uptime: 3 days, 20 mins 
  BBBBBBBBYYYYYY             Shell: bash 5.1.4 
  BBBBBBBBYYYYYYY            Resolution: 1024x768 
  BBBBBBBBBYYYYYW            Terminal: /dev/pts/0 
 GGBBBBBBBYYYYYWWW           CPU: Intel Xeon Silver 4210 (40) @ 3.200GHz 
 GGGBBBBBBBYYWWWWWWWW        GPU: 01:00.1 Matrox Electronics Systems Ltd. MGA G 
 GGGGGGBBBBBBWWWWWWWW        Memory: 6823MiB / 257584MiB

My research is based on electronic structure calculations (Solving the Schrödinger equation for molecules), and I use my nodes in production to carry out demanding calculations.

It was not easy to configure everything as it would be, for eg, in CentOs. But, I must say that I do not regret choosing CL. :slight_smile:

3

Hi @rfkspada , thanks for sharing your view. By saying HPC cluster I mean to run computational tasks across mulitple nodes via MPI or some other protocol. There are serveral benchmark results showing that CL outweigh CentOS or Ubuntu on a single node. It’s almost for sure that it will be better than those distributions on HPC cluster. If there are benchmarks data can show how well it works on HPC cluster comparing with the others, it would be a great help for decision-making.

1 Like