NumLab 2016

---
title: NumLab 2016
author: juha.lento@csc.fi
presentation-date: Spring 2016
layout: reveal
reveal-theme: sky
---

## NumLab 2016

<juha.lento@csc.fi>

### Learning outcomes

-  how to build OpenIFS from source
-  how to run OpenIFS in a supercomputer
-  how to analyse the results

**More** *enjoyable* **and** *productive* **working with computers**

### Pre-requisites

-  general Linux command line skills, working with the directory
   hierarchy and files, editing text files with `vi` or `emacs`.
-  user account in `taito.csc.fi`
-  no prior experience with supercomputers is required

---

## Working with computers

### Golden rules

-  **KIS**, Keep It Simple
-  **DRY**, Don't Repeat Yourself

### Be a programmer, not a computer!

- bash is interpreted language, exactly like python, for example
- use bash like any other interpreted language, write programs
- command line is bash's Read-Execute-Print-Loop (REPL)

### Automate and evolve

- write tasks that you need to repeat or later review as programs
- re-use code, write re-usable code
- learn new tools
- let the machine infer values if possible, do not (re-)set them
  manually
- be efficient with editors and command line

---

## Contemporary supercomputers

### Architecture

![Cluster architecture](./SupercomputerArchitecture.svg "Juha's Incscape art")

Typical contemporary supercomputers consist of *login nodes*,
tightly interconnected *compute nodes*, and accompanying *parallel file
system*.

### Compute node

![Compute node](./ComputeNode.svg "Juha's Incscape art")

Each compute node is a NUMA workstation.

### Parallel file system

![Parallel file system](./ParallelFileSystem.svg "Juha's Incscape art")

Parallel filesystem (Lustre) consists of a MetaData Server (MDS) and
an array of large file servers (Object Data Server, ODS).

### Parallel file system (cont.)

**Keeping thing reasonable**

- avoid excessive metadata access.
- no more than 500 files in a single directory
- single file size < 1GB

### HPC environment vs. workstation

1. multiple versions of software packages
2. more complicated filesystem(s)
3. batch queue system

### Use supers for

-  Parallel computing, massively parallel programs (MPP)
-  farming calculations, parameter searches etc.
-  large data sets

### Further reading

-  <http://en.wikipedia.org/wiki/Parallel_computing>
-  <https://research.csc.fi/csc-guide>
-  <https://research.csc.fi/taito-user-guide>

---

## Questions?