Architecture

Hardware

A cluster hardware is consist of multiple machine with specific tasks i.e. Management, Computation, Storage etc. Arun HPCC is a rather small HPC facility and doesn't have much individual server for each task. Arun HPCC has one master node and three compute nodes.

Master node is responsible to control the entire cluster. It also acts as a login node for external communication, holds all the necessary services and also responsible to control job scheduling.

Three compute nodes runs on minimal linux system so that maximum hardware power can be used for computation tasks. One of the compute node is equipped with GPU device for embarrassingly parallel computation.

Master Node 1 Node 2 Node 3
CPU Intel(R) Core(TM) i5 @ 3.20GHz (4 Cores) Intel(R) Xeon(R) E5-2620 v4 @ 2.10 GHz (16 Cores) Intel(R) Xeon(R) E5-2620 v4 @ 2.10 GHz (16 Cores) Intel(R) Xeon(R) E5-2670 0 @ 2.60 GHz (16 Cores)
RAM 8 GB DDR4 64 GB ECC 64 GB ECC 32 GB ECC
DISK 2 x 250 GB SSD, 1 x 1TB HDD 1 x 250 GB SSD, 2 x 4TB HDD 1 x 250 GB SSD, 2 x 4TB HDD 1 x 1TB HDD
NIC 2 x Gigabit Ethernet 2 x Gigabit Ethernet 2 x Gigabit Ethernet 2 x Gigabit Ethernet
GPU None 1 x NVIDIA Quadro P5000 (16 GB) None None

Software

Whereas hardware for a small cluster like Arun HPCC is rather simple, software architecture on the other hand must have all the necessary components to run a cluster efficiently.

Services Description
SSH Login Control
DHCP Host/IP Management
DNS Domian resolver for cluster
NFS File sharing amount cluster nodes
PSSH Distributed Shell for Cluster Nodes
Slurm Scheduler for controlling computation jobs
Apache Web application for Website and Portal System
Django Backend System for Web Application & User Portal
Gluster Distributed File System for computation data sharing

NOTE

Arun HPCC is a small computation facility. It started with a humble donation of 2 computation nodes from ICTP. And since then we've added few more nodes and designed in a proper cluster configuration.