Page tree
Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 31 Next »

Welcome to the C3 Digital Transformation Institute!

You have been given a grant as part of the new C3 Digital Transformation Institute (DTI)!
To make the start of your DTI experience as fast as possible, we have assembled a set of resources to:

  1. Introduce researchers of all stripes to the C3 system
  2. Help researchers determine what level of training they will need to leverage C3's resources
  3. Point researchers directly to relevant documentation they will need
  4. Provide worked examples of different research workflows and how they may be ported into
    C3's environment, or may use C3's resources

If you have questions not covered by this guide, please contact the DTI team at the email help+c3ai@ncsa.illinois.edu

Introduction to the C3 system

C3 is a Java-based data analytics engine designed to make the ingestion and analysis of heterogeneous data sources
as painless as possible. The C3 system joins data from multiple sources into a single unified federated data image.
With the federated data image defined, C3 then provides an API to access that data, and in the case of time-series data,
perform numerous transformations and computations all producing normalized time-series data at regular intervals.

C3 also supports R and Python Jupyter notebook analysis of the federated data image. These notebooks provide a
great way for researchers to analyze data close to where the data is stored. While C3 supports many data science
capabilities familiar to the researcher, some expected functionality may be missing. For these cases, C3 supports
implementing new data processing functions in python and javascript.

Like any other API porting your own workflows will take some care and time to learn properly. Please leverage this guide
to make understanding C3's platform and porting your workflow as quick and easy as possible.

Services available from C3

  • Covid-19 Datalake: This unified federated Datalake includes data from numerous sources.
  • C3 computing platform
  • C3 Integrated Development Studio
  • C3 Jupyter notebooks
  • C3 Marketplace
  • C3 UI system for creating dashboards

How does C3 differ from traditional HPC systems?

  • Traditional HPC systems are similar to Hardware as a Service (HaaS), while C3 is more like a Platform as a Service (PaaS).
    Users are encouraged to work within the platform's API to achieve the best performance out of C3.
  • C3 offers a state-of-the-art data integration system as the basis for all Data Science operations.
    This is in contrast to HPC systems where all components of data management and the analysis pipeline must be installed and managed independently.

What types of software can be run on C3?

  • Nearly any python module may be installed and used through pip or conda
  • Nearly any R package may be installed and used within the R juptyer environment.

What types of software cannot be run on C3?

  • General binary executables are not supported by C3 out of the box.
  • MPI-based python software
  • Packages which must be built from scratch on the platform, or require specific hardware drivers
  • Python modules which require special built binaries may not run as well.

How do I get started?

Use this guide to determine what training you need to utilize C3's resources effectively. We have separated
researchers into four levels based on what level of interaction with C3's resources they require. We include
basic examples of workflows which might fall into that level, pros and cons of operating on that level, and
a list of training resources we recommend resources researchers completing on the DTI training environment
before starting their C3 allocations. This will ensure researchers will be able to use their allocation as
efficiently as possible.

Examine the high level overviews of each level below, then click the section titles to go to more in-depth
discussions related to that level, like the recommended training.

Level 1: Use COVID-19 Datalake Only

For many researchers, they will simply want to leverage the C3 COVID-19 Federated Data Image.

Pros:

  • Easy to integrate into existing scientific workflows and run on existing scientific computational hardware
  • Publicly available API means no credentials are needed to access the data
  • Assuming you have access to your own computational resources, you don't have to worry about allocations
    on C3's platform.

Cons:

  • All data used from the Datalake must be streamed to wherever you're processing data
  • Performance benefits from working with the Datalake using C3 will not be available.

Level 2: GUI based data analysis on C3

C3 provides a wonderful GUI based interfaced to the C3 system with their Integrated Development Studio. Such
an environment is likely to be attractive to many researchers. This level is the easiest way to integrate new
data onto the Datalake.

Pros:

  • GUI interface to manage C3 Types and data integration.
  • GUI interface to piece together ML pipelines.
  • Ability to load new data onto the Datalake

Cons:

  • Some types of workflows may not be easily defined within the GUI framework

Level 3: Utilize C3 AI Suite and Jupyter notebook analysis

Some researchers will want to write their own C3 package and leverage more of the AI Suite through Jupyter notebooks.
C3 allows researchers to define their own types, methods, and use R and python to perform analysis on Datalake data.

Pros:

  • Researchers can use a jupyter notebook to interface directly with their Data Model.
  • Researchers can often use exactly the same workflow they were using before.

Cons:

  • We recommend users take a full set of training to completely familiarize themselves
    with the C3 system before embarking on their analysis.

Level 4: State-of-the-art ML workflows requiring special ML models and/or GPUs

Some researchers will want to bring state-of-the-art ML workflows to C3. C3 can support such workflows, but
extra work may be needed.

Pros:

  • Researchers can bring state of the art workflows close to the COVID-19 Datalake

Cons:

  • The DTI team will evaluate on a case-by-case basis whether a workflow is appropriate for C3.
  • Some workflows may require major effort to fit within the C3 framework.

Accessing C3

This section introduces the process to access C3. Generally speaking, once you receive your grant,
the DTI team will reach out and discuss with you what your needs are.

  1. Determine which researchers will require access to a C3 tag.
  2. Each researcher will be given a C3.ai developer portal login.
  3. Each researcher will be given a tag on the C3.ai DTI training cluster.
  4. Once training is complete, Discuss with the DTI team what your needs
    for a C3 cluster will be.
  5. The C3.ai DTI will work with C3.ai to stand up a new tag for your research.
  6. Access to that tag will be granted to your researchers
  7. Research can then proceed until your allocation is exhausted!

Essential C3 Concepts

C3 is quite different from traditional HPC resources. We have written an introduction to C3 from the
perspective of a scientific researcher. We go over several important C3 concepts and relate them to
what scientists are more familiar with.

C3 Allocation Management

This section introduces How researchers will be expected to manage their allocation while on the C3 platform.

This section will be expanded once the DTI team understands how this procedure will look to the researcher.

Special Compute Resource Information

Here you can find information about the special compute resources available to C3 DTI researchers.

Comprehensive List of Available Training and Resources

See the above link for a comprehensive list and categorization of the available training
materials. This includes C3 Documentation, DTI introductions, and DTI created examples and exercises.

Help! This guide doesn't solve my problem!

No problem! You're not alone! Please send an email to help+c3ai@ncsa.illinois.edu with a description of your issue
and one of our team will work with you to resolve your issue.

Feedback

If you feel aspects of this guide are incomplete or Inaccurate, please send an email to help+c3ai@ncsa.illinois.edu with the
issue or suggestion, and we will work to incorporate it to make the documentation better. We appreciate the new perspective
More eyes can bring to a software project!

Your DTI Team

NCSA

Jay Roloff - Executive Director

Matthew Krafczyk - Data Analyst

Yifang Zhang - Data Analyst

Berkeley

Larry Rohrbach - Executive Director

Eric Fraser

Greg Merritt

Matt Podolsky

  • No labels