ALCF Data Science Program: Proposal Instructions

 

The ADSP Proposal Process

The ADSP projects will be categorized as either "data science projects," which will have a specific science goal, or "software framework projects," which will be focused on implementation or development of a specific framework required to support data science, including complex workflows. Crosscutting proposals which combine the two are encouraged. Projects will need to self identify in their proposals.

Proposals must have a clear plan for the science to be accomplished or framework to be implemented. Proposals should include a detailed timeline describing the application development or science application runs which will occur throughout the duration of the project.

The forms and instructions attached should include everything needed to submit a proposal. These are, roughly, a simplified version of an INCITE proposal. Please direct any questions to adsp@alcf.anl.gov.

Proposal deadline is Wednesday, June 20, 2018, 5:00 PM CST.

Evaluation of Proposals

ALCF staff, with the help of internal and external experts, will evaluate proposals on the strength of:

  • Contribution to state-of-the-art of proposed science or software framework.
  • Potential impact of proposed science or software technology.
  • Use of and need for the architectural features of Theta (e.g., node-level parallelism and the memory hierarchy on the KNL nodes).
  • Scalability of the target application(s) on Theta and/or Mira. Applications that have not run on these systems should describe scaling studies on other systems, and the effort required to successfully migrate to ALCF systems.
  • Description of the data-scale requirements and plans to realize the data science and/or technology at these scales.
  • Appropriateness of development team: is it likely that project member expertise and person-hours proposed are likely to accomplish the science goals or software developments.
  • Overall diversity of science domains and algorithms.
  • Prospects as an Aurora application, and intent to use Aurora.

Submission

  • Submission deadline: June 20, 2018 at 5:00 PM CST
  • We are using the EasyChair system for proposal submission. You will need to create an account if you don’t have one already. Then log in to the Argonne Data Science Program EasyChair website.
  • Prepare your proposal using the instructions below
  • Submit as a single PDF document, by using EasyChair to upload. You may resubmit with revisions as needed up until the deadline.
  • Please direct any questions to adsp@alcf.anl.gov.

Proposal Instructions

Please create your proposal document with a project title, and the section headings noted below:

Proposal Title: Please include a project title

Section 1: PI and co-PI Information

1a. Principal Investigator (PI) Information

  • Last Name, First Name, Title (Dr., Mr., Ms., etc.)
  • Institution
  • Street address
  • Email address

1b. Co-Principal Investigator (co-PI) Information

For each co-investigator:

  • Last name, first name, title (Dr., Mr., Ms., etc.)
  • Institution
  • Street address
  • Email address

Section 2: Project Summary

2a. Executive Summary

  • Write an executive summary that accurately describes your proposed research and the high-impact scientific advances you will achieve with access to resources at the ALCF.  (1/2 page). 

2b. Benefit to Community

  • Write a description of the benefit your project will provide to the science community (1/2 page).

2c. Impact Statement

  • Provide a two-sentence project summary that can be used to describe the impact of your project to the public (50 words maximum).

2d. Science/Software Framework Summary

  • Please identify the category: Science/Software Framework/Both
  • If submitting as a science project: Write a description of the science problem you would like to address throughout the award. Include research that will need to be completed in the next two years to achieve these results (1 page).
  • If submitting as a software framework project: Write a description of the framework you would like to develop and deploy within the award periodInclude any research or development that must be completed to build up to this work (1 page).

2e. Application Summary

  • 2e.i. Application Requirements: Write a list of your application requirements, including languages (C, C++, Fortan, Java, Python, etc.), toolkits and frameworks (Tensorflow, Neon, Torch, Caffe, etc.), libraries, and current parallel method (MPI, OpenMP, SPARK, etc.) (1 page).
  • 2e.ii. Application Description: Write a description of the current application, including methods, parallelization, workflows, I/O, etc. (1 page).
  • 2e.iii. Application Performance: Describe any performance data, such as scaling, of the current application components, including methods, I/O, workflows, etc. (1/2 page).
  • 2e.iv. Application Development Needed: Describe your intent to use architectural features of the ALCF machines; for example, node-level parallelism and the memory hierarchy on Theta. Consider how you might use the SSDs on each node of Theta for modalities including but not limited to data persistence, data staging, out-of-core accesses (1 page).
  • 2e.v. Application Data Requirements: Write a description of your data management requirements. Example topics to include:
    • Streaming/real-time data feeds
    • Access to remote databases 
    • Setup/access of local databases 
    • Data formats (HDF5, etc.) 
    • Data sizes/scales (ingested/output data for each stage in end-to-end workflow)
    • ALCF persistant storage requirements (number of bytes, maximum number of files, other measurements that pertain to non-file-based storage)

Section 3: Estimate of Resources Requested

In this section, please describe the hardware resources required for the planned work. Projects may request time on ALCF resources:

  • Theta has 4,392 nodes, each with a KNL 64-core processor having up to 16 gigabytes of high-bandwidth in-package memory and 192 gigabytes of DDR4 RAM. Each node has a 128GB node-local SSD storage. The aggregate peak compute speed is 11.69 petaflops. It has a 10 petabytes Lustre parallel file system.
  • Mira is a 10-petaflops IBM Blue Gene/Q system consisting on 48K nodes with a 5D torus interconnect; each node has 16 cores with four hardware threads per core and 16 gigabytes of RAM.
  • Cooley is a visualization and analytics cluster with 126 compute nodes; each node has 12 CPU cores and one NVIDIA Tesla K80 dual-GPU card. The entire Cooley system has a total of 47 terabytes of system RAM and 3 terabytes of GPU RAM.

3a. Theta Resources:

  • Theta time in core-hours
  • SSD Use
  • Storage in TB
  • Tape archive space in TB
  • Network Requirements
  • Breakdown for how you would use time on Theta to make final preparations for science runs, and for the science runs themselves. Preparations might include final scaling tests, science problem spin-up runs, etc. For the science runs themselves, estimate the total core-hours and break down into separate components/milestones as appropriate. You will have access to computational resources for the two-year period. (1/2-1 page).

3b. Mira or Cooley Resources:

  • Mira/Cooley time in core-hours
  • Storage space in TB
  • Tape archive space in TB
  • Network requirements
  • Brief schedule for how you would use that time on Mira and Cooley: scaling tests, development (e.g., algorithms), verification, parameter sweeps, etc. Break this down into milestones as appropriate for your project (1/2 page).

Section 4: Other Collaborations

Indicate whether your team, or others you are aware of using the same code base, have projects using other large-scale resources such as NERSC, OLCF, etc.

Section 5: Project Team Members

5a. Names and Levels of Effort

  • List the names and levels of effort (as a percentage of full-time) for all team members you expect to do work on the ADSP project.
  • For each person, include a CV. If you have trouble getting all of the CVs into the PDF proposal document you are submitting email adsp@alcf.anl.gov for assistance.
  • We may fund projects, in part, with postdoctoral scholars working on methods in data science as well as on data-centric runtimes/infrastructures. The postdoctoral scholars will be shared with other ADSP projects. Please identify the tasks you would like the postdocs to contribute to your project.