OpenMined PipelineDP

Attributes

Description

Informational

Type

New

Industry Sector

Data Centers and Data Warehouses of Hospitals, Govs, Teleco and Schools

Business driver

Differentially private data aggregation framework, pipelineDP, enables write fast, flexible pipelines that use modern techniques to aggregate user data in a privacy-preserving manner.

PipelineDP is a framework for applying differentially private aggregations to large datasets using batch processing systems such as Apache Spark, Apache Beam, and more.

To make differential privacy accessible to non-experts, PipelineDP:

• Provides a convenient API familiar to Spark or Beam developers.
• Encapsulates the complexities of differential privacy, such as:
  • protecting outliers and rare categories,
  • generating safe noise,
  • privacy budget accounting.
• Supports many standard computations, such as count, sum, and average.

Business use cases

1.Schools do machine learning on differentially private data aggregation infrastructure for confidential student information.

2. Hospitals do machine learning on differentially private data aggregation infrastructure for confidential patient information.

3.Gov do machine learning on differentially private data aggregation infrastructure for confidential wage information.

4.Telco do machine learning on differentially private data aggregation infrastructure for confidential consumer text msg and phone call records information.

Business Cost - Initial Build Cost Target Objective

Edge Cloud should be deployable with more than 3 servers in a single rack at a low cost.

Business Cost – Target Operational Objective

1. execute this framework on top of disaggregated datasets in Data Centers and Data Warehouses requires little cost.
2. In-place upgrade of the Data Centers and Data Warehouse should be supported without impacting the availability of the edge applications
3. The automation should also support zero touch provisioning and management tools to keep operational cost lower

Security need

The solution should have granular access control and should support periodic scanning.

Regulations

The Edge cloud solution should meet all the industry regulations of data privacy and telco standards (NEBS).

Other restrictions

Consider the power restrictions of  specific location in the design (example - Customer premise, where data are stored in School's internal servers)

Additional details

The Edge Cloud Solution should be deployable across the globe and should be able to support more than 10,000 locations.

 Use case submitters can include  SQL queries get/set.

Case Attributes

Description

Type

New

Blueprint Family - Proposed Name

Differentially private data aggregation framework – OpenMined PipelineDP

Use Case

Differentially private data aggregation

Blueprint proposed Name

Differentially private data aggregation framework: OpenMined PipelineDP

Initial POD Cost (capex)

Unicycle less than $150k: 3 Arm bare metal machines, 1 10G switch

Scale & Type

For the smallest deployment, this requires 2 Arm bare metal machines. For large deployments, this could span to large number of bare metal machines.

Applications

Differentially private data aggregation for large scale online education, telemedicine, Hospitals, Govs, Teleco and Schools.

Power Restrictions

N/A

Infrastructure orchestration

Host:

•Orchestrator: Kubernetes

•Bare Metal Provisioning：Ansible

•Kubernetes Provisioning：KuD

•OS: Ubuntu

•GPU Driver: AMD,NVIDIA:

•Network: OVS

•GPU Driver (AMD, NVIDIA)

SDN

N/A

Workload Type

•Data Center SQL databases

Here are some examples of how to use PipelineDP:

• Apache Spark example
• Apache Beam example
• Framework-free example
• Example with all frameworks

Additional Details

N/A