CISnapshot

From NJIT-ARCS HPC Wiki
Jump to: navigation, search

HPC Clusters

As part of the Tartan HPC Initiative NJIT supports two HPC clusters, with a combined capacity of:

  • CPU
    • 379 nodes
    • 3,568 cores
    • 30,976 GB RAM
  • GPU
    • 24 nodes
    • 76.248 cores
    • 280 GB RAM

    Details: HPC Extract, HPC specs

  • Theoretical aggregate of approximately 55 TFLOPS, exclusive of GPU
  • GPU types: Nvidia K20/K20x/K20m, Nvidia Tesla Pascal 100; Nvidia Volta V100 being considered for Computer Science Data Science program.
  • Most CPU nodes (314) have Nehalem processors, circa 2009. These nodes were obtained as a donation from Linode. CPU node RAM ranges from 64 GB (most common) to 256 GB. Recent additions to the clusters have Westmere, Sandy Bridge, Ivy Bridge, Haswell and Broadwell processors.
  • The vast majority of nodes are connected with gigE; other interconnects are 10gigE (20) and FDR InfiniBand (42).

Migration to OpenHPC

Both HPC clusters are scheduled to migrate to the OprnHPC model] in the coming months.

For that migration, a development cluster has been set up for testing the installation and deployment of the components of the OpenHPC software stack. It is expected that this cluster, or a similar one, will become a testbed for researchers.

The OpenHPC scheduler/workload manager is SLURM. The SLURM project is very well-supported and active. SLURM will replace the Grid Engine scheduler/workload manager currently in use on both HPC clusters.

OpenHPC is expected to be implemented on the smaller of the HPC clusters in Spring 2019, and on the entire (unified) HPC cluster in Summer 2019.

Hadoop/SPARK cluster

    2 x IBM iDataPlex dx360 M4 nodes, each with:
    • 2 x Intel Xeon CPU E5-2680 (8 Core)
    • 16 CPU CORES @ 2.70GHz
    • 32 Logical Processors with Hyperthreading
    • 128G RAM

Data Center for Hardware Owned by Researchers

GITC 4320, in pods: 13 physical machines, owned by 5 research groups, in 5 pods

GITC 4320, not in a pod: 1 cluster, owned by 1 research group; 1 server, owned by 1 research group

Details: GITC 4320

Storage

Research groups are allocated 500GB of both AFS and NFS storage.

Additional storage can be purchased. Cost schedule.

Various storage capacities

Researcher Problem Domains

Researcher Problem Domains

Software

A wide spectrum of open source and commercial software, compilers, libraries, and utilities is available on all HPC computational resources, much of it via the AFS distributed file system. Software Categories

Details: Software modules