#presentation.Rmd#

---
title: "Parallelization and Remote Servers"
author: Abby Bratt, Maria Kuruvilla, and Colin Okasaki
date: February 5, 2020
output: ioslides_presentation
---

# Why Parallelize?

## Big Data and Slow Code

- Lots of data, simple operations, or:
- Not so much data, complex operations, or even:
- Lots of data, complex operations

## (Dis)Advantages of parallelization

- $N$ cores may speed up your code by a factor of $N$, but:
  - only if you can efficiently distribute your code to cores
- Iterative algorithms: the bane of parallel computation
- Communication: the real bane of parallel computation

## Basic Structure of a computer

- Some CPUs (processors)
  - Each CPU has some cores
  - Each core can do 1 computation at a time
  - Theoretically top speed = all cores active
  - Reality: cores need to communicate

## Why use a remote server

- More cores
- Built for heavy computation
- Doesn't slow down your personal computer
- Feel like a pro, brag to your friends

## Recap

- When should you parallelize
- When should you not parallelize
- When should you use a remote server

## Practice

- MCMC
  - Yes and no
  - Can run multiple chains but each chain is iterative
  - Computationally intensive, run remotely
- Bootstrap
  - Yes
  - Embarassingly parallel
  - 
- Likelihood evaluation
  - Often, for big, conditionally independent data
- Linear algebra
  - Yes, for some operations
  - Matrix multiplication, e.g.
- Optimization

# How to Access the QERM/SEFS Servers

# A Basic R Program

# How to Access Hyak