CS 5220: Applications of Parallel Computers

Objectives

• Reason about code performance
  • Many factors: hardware, software, algorithms
  • Want simple, "good enough" models
• Read/judge HPC literature
• Apply model numerical HPC patterns
• Tune existing codes for modern HW
• Apply good software practices

Prereqs

Guided by basic logistical constraints:

• Default class codes will be in C
• Our focus is numerical codes

Fine if you're not a numerical C hacker!

• I want a diverse class
• Most will have some holes
• Come to us if you have concerns

Usual Logistics

• Home: cornell-cs5220-f15.github.io
• Read the syllabus and conduct code
• Enroll if you'll take the class
  • If you're note allowed fill the waitlist survey
  • Regular registration should be open Aug 26
• Grading breakdown
  • Class participation + HW (20%)
  • Standard group projects + peer review (50%)
  • Final project (30%)

Unusual Logistics

• Intel Xeon Phi cluster
• Git-oriented workflows
• Public development work
• Flipped classroom (bring laptops!)

More on all of this in a bit!

How many are you?

Let's try the following simple counting protocol:

• Rightmost person counts number in each row
• When done counting, raise hand
• State number when called on

Estimate how long should this take. Then let's see!

How many are you?

Would it be more efficient for me to count?

• Time to start the process?
• Time to compute vs communicate?
• Opportunity for parallelism?
• Balanced work partition?

Who are you?

Icebreaker: Greet your neighbors

Who are you?

How many have or are pursuing

• A CS degree?
• Engineering degree (outside CS)?
• Other science degree (outside CS)?
• Other degrees?

Who are you?

How many are

• Undergraduates?
• Masters students?
• PhD students?
• Other?

HW0: Who are you?

• Create a GitHub account
• Fill out Assignment 0:
  • Netid and GitHub account
  • Demographic information
  • Why taking the course
• Submit via CMS

Logistics: Accounts

You'll need two accounts:

• Cluster account (netid login)
  • I'll start with facultycenter enrollment
  • Help me out: enroll or send me email!
• GitHub account (may be pseudonymous)

Xeon Phi cluster

• Fancy HW toy courtesy Intel + Dell discount
  • Eight 12-core host nodes
  • Fifteen 60-core Xeon Phi accelerators
• Typical HPC software environment
  • Linux base + environment modules
  • Submission via batch queue

A new cluster -- expect issues, be patient!

Git

• Popular Distributed Version Control System (DVCS)
• You'll learn to use it in this class
• Why?
  • A good standard tool
  • Helps with teamwork
  • Helps with multiple machines

See also: "Rationale for class Git workflows"

Git + GitHub

• Fork/pull: very common software workflow
• Start assignments by forking a class repository
• Submit work via pull requests (PR)
  • Start a conversation (get feedback!)
  • Can keep working after PR

Course correction/addition PRs for extra credit!

GitHub

• Default for distributing work and submitting
  • Public code = bragging rights
  • You learn from your peers
• Privacy concerns
  • Respect the privacy of your peers!
  • Grades and such stay on CMS
  • May stay pseudonymous (use private email)
  • Can opt out without penalty -- come talk to me

Flipping

• Lectures and reading out of class
• Class time for collaboration and interaction
  • Discussions of readings
  • Pair programming
  • Guided problem work
• Bring a laptop if you've got one!

Out of class

• Class preparation work
  • Readings + responses, short exercises, etc
  • You may update responses during/after class
• Group projects
  • Parallelization / tuning of existing codes
  • Performance analysis is a big component
  • Groups assigned to mix background
• Final project
  • Research track: write a paper
  • Teaching track: develop a module

Projects

• Deliverables
  • Tuned (and correct) code
  • Written report with reproducible experiments
• Three or so during semester
• Assigned small groups
• Comes with milestones
• Work is peer reviewed
• Groups also self-assess

Sharing

• Default: class code (including yours) is open
  • Goal: study and improve peer codes
  • Caveat: cite what inspires you
  • Bulk copying with citation still scores low
• Peer review partly to force sharing

Late work

• In-class work and peer review need timely work
• Two parts to grade: initial and final submission
  • Get feedback on initial submission
  • Correct issues before final submission
• PRs track your repository updates

Final projects

• Peer-reviewed research or instructional projects
  • Small groups (unassigned)
  • Will solicit proposals in a month or so
• Research projects
  • Short research paper and presentation
  • Topic must involve performance
• Teaching projects
  • Instructional module and presentation
  • Includes learning goals, assessment, deployment plan

Marching orders

• Home: cornell-cs5220-f15.github.io
• Read the syllabus and conduct code
• Create a GitHub account
• Submit HW0 on CMS
• Read software setup recommendations
• Read performance analysis notes
• Come Thursday with computers!