IBandPFS

Roles of high-speed internal network and parallel file system

A high-speed node interconnect - e.g., InfiniBand (IB) - and a parallel file system (PFS) are both standard components of an HPC cluster. Both are proposed for the HPC-BR.

Together, compared to the current internal network and filesystem used in the HPC clusters, they would vastly increase perfromance for a wide class of computations typically run on the NJIT clusters.

High-speed internal network

The HPC internal network is used for communications between compute nodes, and also between compute nodes and data storage.

The current Kong cluster supports an internal network with 1Gb/sec capacity, except for 25 dedicated nodes, which have a capacity of up to 56Gb/sec.

The proposed HPC-BR specifies an internal network capacity of 100Gb/sec for all general-access nodes.

Parallel file system

The HPC cluster model uses hundreds of compute nodes, each containing several CPUs, each of which contains several processors (cores) to perform calculations. Jobs running on these cores in general write and read files to/from disk as part of the computational processes, and are essential to the processes.

When the temporary files are more than a few gigabytes in size, the most time required for computations is consumed by writing to and reading from files on disk - i.e., disk I/O is the bottleneck. In addition, parallel jobs handling files of any size will encounter this bottleneck.

The proposed HPC-BR specifies a 1-PB PFS.

The condominium model already in use for Kong, in which researchers can purchase a dedicated portion of the hardware, can also be employed for the PFS appliance.

Problem domains that will benefit from high-speed interconnect and PFS

The following problem domains for which the HPC clusters are currently used will benefit from the proposed HPC-BR high-speed interconnect and PFS:

• Bioinformatics
• Clustering analysis and data mining
• Computational acoustics
• Computational algorithms
• Computational biophysics
• Computational chemistry
• Computational fluid dynamics
• Computational partial dufferential equations
• Computational physics
• Condensed matter physicas
• Granular matter physics
• Cryotograhy
• Data science
• Image processing and 3d visualization
• Materials research
• Monte Carlo simulation
• Neural networks, geneteic algorithms
• Software verification
• Solar-Terrestrial research
• Statistical analysis