Welcome to Arun HPCC documentation and usage guidelines. This documentation is designed to help you to work with Arun HPCC. Like any High Performance Computing facility Arun HPCC is a remotely accessible facility which is shared among many users for their high performance computing need. Being a shared facility it is very important to be responsible while using the facility. The general rule is Do not do anything that can or might cause problem to other users. Let's get to know Arun HPCC first.
A HPC facility is consist of many nodes. Nodes are typically separated in two main category.
- Head Node
- Compute Nodes
Head Node or nodes, which is also sometimes called control nodes are responsible to manage various aspect of the
cluster facility. This also acts as a
Gateway for the cluster to the world. Head node(s) also controls the user login
and the job submission to the cluster. User basically login into the
Head Node via
SSH over the internet. After
logging in the user can create and submit
Jobs for processing. Head node(s) also controls the computing resources,
their allocation and running the user
On the other hand
Compute Nodes are responsible to run the submitted user
Jobs. Head node allocate appropriate
hardware resources from the compute nodes according to the
Job request. Compute nodes follows the instruction of the
Job script and process the computation.
Recent days GPU is no longer for gaming only, it has becomes a major hardware for massively parallel computation. It is called GPGPU or General Purpose GPU. From embarrassingly parallel scientific code to tackle huge computation of an Artificial Neural Network GPU is being using everwhere.
Arun HPCC is not far behind. It has one computation node with Nvidia Quadro P5000 GPU. With this GPU node Arun HPCC also steps into the massively parallel world of GPGPU.
Head Node and
Compute Nodes big cluster has separate nodes for management of differenct resources. E.g.
Arun HPCC is a small computation facility. So it has one
Head Node which can easily handle the management task of the
cluster as well as handle the job scheduler.