Littlefe-workshop

From Education

Jump to: navigation, search

Contents

The Proposal

  • See cvs:cluster.earlham.edu:/LittleFe/IntelWorkshop/proposal.tex for the final submitted proposal.

Proposal Notes

  • A 4 day workshop designed to teach teachers how to include multi-core, many-core and distributed computing topics into the undergraduate CS curriculum. The materials are centered around LittleFe and the Bootable Cluster CD, both of which are specifically designed for teaching parallel and distributed computing.
  • 1-2 faculty and optionally 1-3 students per institution, one LittleFe per institution

Day 1

  • Introduction and overview
  • Assembling and testing LittleFe

Day 2

  • Installing and configuring the BCCD
  • Shared memory parallelism with OpenMP

Day 3

  • Distributed memory parallelism with MPI

Day 4

  • GPGPU parallelism with CUDA
  • Hybrid parallel models, MPI + CUDA

Budget

  • $100K total
  • Stipends for instructors and assistant instructors
  • Hardware for the participants
  • Room and board
  • Participants pay a relatively modest fee (say $500) and their own transportation

Telecon on Nov 5th

  • Wilf, Kimberly, Tom, Charlie, PaulS
  • Workshop idea
    • Wilf's caveat, truly portable so that we can get material out broadly, scaling, captured and posted
    • Scaling built-in to the plan, not at workshop necessarily but as part of the project plan, maybe a workshop just for the scaling
  • Wilf is willing to talk to SC12 chair (JeffH) and Education Program chair (unknown)
  • SC might be willing to help with funding (Wilf to consider/talk)
  • Wilf, this is a seed, Paul/Wilf/Michael willing to help with funding going forward if we can make this first bit successful
  • Do the workshop for filming, etc. during the summer of '11 so that it's all in-place by the SC11 workshop
  • Kimberly would like to target other universities as well of the ones on her list
  • Kimberly would like us to build a couple of systems for them as sandboxes
  • Applicants should put a proposal in for why they want one and what they'll do (in addition to the fee)
  • Better name than LittleFe? Box-o-cluster? Cluster-in-a-box?
  • If someone gets a system and they develop materials and contribute back there is some mechanism for rewarding them (BlackBelt)?
  • OpenCL rather than CUDA, Wilf thinks TimM and others might be supportive of this. NVIDIA supports this? Check!
    • Wilf will check with TimM about materials, etc.
  • Kimberly/Wilf might be able to leverage Knightsferry and Knightsbridge in the future.
  • Charlie will check with Shodor about being a fiduciary agent.
  • Nov 12th due, Nov 19th cutoff for funds
  • Kimberly will check with her grants people about ACM and Shodor being fiduciary agents

Questions for K

  • Ask about Wilf conversation
  • Ask about SC11 and SC12 coordination as site to leverage people being there and to have suitably sized conference space, good synergy with SC Ed, who may be able to underwrite space
  • How many people would they reach with $100K (we are currently targeting ~60 people and 15 institutions)
  • Can we re-distribute the Intel toolchain?
  • Do they have any religion about CUDA? (In a sense I'd rather not raise this, lets just let it be on the board and silently available)
  • Who would we talk to at Intel about hardware donations?

Questions

  • How can we work with EAPF? What do we have to offer each other?
  • Can we leverage hardware donations? Doubtful since we need specific motherboards and not just CPUs.
  • Is a 5 day model (one day for GPGPU, one for hybrid and debugging) better?
  • Is a 3 day model (one LF/BCCD, one shared, one distributed) better? Lower cost = more workshops = more exposure
  • Can we re-work all our materials to use the Intel toolchain?
  • Make it clear that we will provide on-going support (hardware and software) after the workshop.
  • What does Michael mean by "lots of exposure" and "not like METU"?
  • Recruit from Intel's school list but don't try to hold the workshops there

Tom's notes

design constraints

  • What a participant walks home with
    • LittleFe they built
    • One per institution (we recommend 2-3 faculty and 1-3 students attend per institution?)
  • perceived value of LittleFe
    • teaching parallel CS
    • ability to work with cloud computing
    • clearly seen useful for institutions with no parallel infrastructure
    • also useful even with well endowed institutions as device student team can use and crash without harming others
    • sustainability is important, EAPF may supply us this
  • Number of participants
    • we said we can scale beyond 30
    • lets target 1-3 faculty and 1-3 students attend per institution,
    • means we need auditorium, rather than just large classroom, for plenary meeting
    • can (prefer) have breakown rooms, one team per corner or room
    • not clear if this translates to requirement for more than one 20amp circuit* back of envelope current budget
  • registration fee
    • All agree needs to exist
    • can't be onerous and drive people away
    • $500-$1000 reasonable per institution since getting a $3K LittleFe
    • Perhaps $100 per person
  • site
    • Michael made it clear we are not doing our summer workshop model (aka METU model)
    • even if that is what we are really doing (since all agree a face-to-face workshop is right thing
  • $$
    • proposal for $75-100K
    • desire to stretch funds over two years (means can write it to presume EAPF ongoing funding)
    • needs to be registration fee
    • May be able to get donation of Atom chips, not clear if there is more, does this mean nVidia board is also possible donation?
  • back of envelope budget
    • $75K first year , $25K second year with $50K fill-in from EAPF
    • $4K to each of Earlham and CCC to develop version of LittleFe used with this program
    • 15 institutions at workshop so $45K towards hardware
    • 4 day workshop, 60 people (4 people per institution), $125/per person/day housing and foods cost -> $30K
    • $16K in travel costs for us and six shoulder surfers
    • $1K stipend for us, $500 for students -> $5K
    • ($20K) should come in from registrations
  • other stuff
    • must be US focused
    • We will be getting list of Universities at which Intel recruits


Objects currently visible floating on the churn

Charlie's notes

  • Wilf now manages a group of 9w+1m technical people focused on ??? University Programs???
  • Intel would like a learning sandbox environment where students can work.
  • Attaching workshops to conferences, this is Intel's push. Michael is now in a group with "scaling" in the name so having more exposure is a good thing.
  • Kimberly stepping in for Scott while he is on sabbatical for a couple of months.
  • 36 universities in the US that Intel focuses on, Kimberly will send us this list
  • $75-$100K for this. Can we find matching funds? Let's go for $100K, let them talk us down.
  • Call while in NOLA, Kimberly won't be there but we can organize a call.
  • Deadline for the proposal is November 12th.
  • Work backwards from the $100K number with the proposal including that we can get hardware from Intel and possibly others. Chip donations are a possibility, Atoms.
  • 1 LittleFe to 1 person, or 1 LittleFe to 1 institution mapping.
  • Setup a program that's like a club membership, order the parts and then do it themselves. 2011 to figure-out the details and then offer it more broadly.
  • The tech people are all HPC, What about cloud systems? What about LF as a gaming system?
  • From Intel's perspective, what's the best possible outcome?
    • US, not so much international (beveryly or someone else later)
    • Universities where they do lots of recruiting (Kimberly sending a list)
    • If it's possible to charge a fee, that would help stretch the grant from Intel, on the other hand what does Intel out of this? More schools teaching parallelism. Real fee is committment to going forward with this. Intel working on rewards program for building and using multi/many core systems in their classrooms.
    • This is about scale, how to make it as big as possible
    • This is likely a one-shot deal for 2010, might be possible to do it again but not likely.
  • Kimberly will talk to Wilf about soliciting other companies.
  • EAPF is the perfect vehicle for this.
  • We need to organize our on-line curriculum materials, they are spread-out all over hither and yon.
Personal tools
SC Education sites