Internet-Draft DAP DP Extensions October 2024
Thomson Expires 21 April 2025 [Page]
Workgroup:
Privacy Preserving Measurement
Internet-Draft:
draft-thomson-ppm-dap-dp-ext-latest
Published:
Intended Status:
Standards Track
Expires:
Author:
M. Thomson
Mozilla

Distributed Aggregation Protocol (DAP) Extensions for Improved Application of Differential Privacy

Abstract

The Distributed Aggregation Protocol (DAP) can be a key component of a system that provides differentially-private guarantees for participants. Extensions to DAP are defined to support these guarantees. This includes bindings of reports to specific options, so that the aggregation service can better implement privacy budgeting and replay protections.

About This Document

This note is to be removed before publishing as an RFC.

The latest revision of this draft can be found at https://martinthomson.github.io/dap-dp-ext/draft-thomson-ppm-dap-dp-ext.html. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-thomson-ppm-dap-dp-ext/.

Discussion of this document takes place on the Privacy Preserving Measurement Working Group mailing list (mailto:ppm@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/ppm/. Subscribe at https://www.ietf.org/mailman/listinfo/ppm/.

Source for this draft and an issue tracker can be found at https://github.com/martinthomson/dap-dp-ext.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on 21 April 2025.

Table of Contents

1. Introduction

The Distributed Aggregation Protocol (DAP) [DAP] can be used as part of a differentially-private system.

Differential privacy depends on being able to limit the contributions from participants. The basic mechanism that DAP uses to cap contributions is anti-replay. Aggregators are responsible for ensuring that the same report cannot be aggregated more than once. An honest participant will contribute a limited number of reports and can rely on at least one aggregator preventing that report from being used multiple times. (The threat model does not seek to protect the privacy of a dishonest participant.)

This basic anti-replay mechanism allows DAP to provide caps on contributions. The resulting system is somewhat inflexible, which can limit the applicability of the protocol outside of the narrowly-defined usage modes in the basic specification.

This document defines several report extensions to DAP that either enable greater flexibility or help constrain the flexibility allowed by other options.

2. Conventions and Definitions

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

3. Late Task Binding

DAP presently requires that a client be aware of the task that it is contributing to. The identity of the task is bound to each report through the inclusion of the task ID in the call to the sharding function of the VDAF (Section 5.1 of [VDAF]).

The late_binding report extension (codepoint 0xTBD) signals to aggregators that a report was not bound to a specific task when it was created.

Late task binding might be useful when reports are collected by an intermediary. The client that generates the report in this case might be unaware of how the report will ultimately be aggregated. This allows the intermediary to defer the creation of a task until it has determined the necessary parameters for the task.

When sharding and protecting reports, the task ID is replaced with a the fixed, 32-byte sequence of b13e8440f1cdb4da51eed3967e0a2652d27f5005bc35f751daf188b4b746708b (in hex).

Enforcing anti-replay for a report that is not bound to a specific task is challenging. An aggregator cannot constrain its search for duplicate reports to those that were submitted to the task. This could greatly increase the cost of meeting anti-replay requirements. The intent with this extension is that additional constraints, such as one or more of the scoping extensions (see Section 4), will be used to make it more feasible for an aggregator to comply with anti-replay requirements.

4. Scoping Extensions

The DAP report extensions in this section might be used to either constrain the use of reports for tasks that are configured with matching values or group reports for the purposes of detecting duplicates.

Including additional scoping information can also ensure that reports do not get reused outside of their intended scope.

This section defines report extensions that carry requester identity (Section 4.1) and report partition (Section 4.2).

4.1. Requester (Website) Identity

Reports might be requested by an entity that operates at lower trust level than the entity that assembles the report. The entity at the lower trust level might not have access to the information necessary to generate the report.

The requester_identity report extension (codepoint 0xTBD) contains an encoding of the entity that requested the report be created.

For example, an application could ask the operating system to generate a report using information that would normally be withheld from it. Similarly, a website could ask a web browser to generate a report based on otherwise secret information. In either case, the release of information for report is conditional on it only being used by a specific aggregation service under terms that have been previously established with the aggregators. Binding the report to the identity of the requester ensures that any use of the system can be accounted for as coming from that requester.

The specific encoding used in this extension will depend on the application. However, the use of a globally-unique identifier, such as an origin ([ORIGIN]) or serialized site ([SITE]), reduces the likelihood of name collisions. A name collision might either allow two requesters that share an aggregator to share and reuse each others reports (or perhaps to marginally increase the odds of having reports spuriously detected duplicates).

4.2. Report Partition

This extension allows a client to bind a report to an application-defined label. This allows applications to partition reports and have each partition managed separately.

The report_partition report extension (codepoint 0xTBD) contains an application-defined sequence of bytes.

The use of this report extension allows aggregators to partition their state for tracking reports. Duplicate reports only need to be tracked across a matching partition, for detecting duplicates within a task or for detecting duplicates across tasks.

The selection of partition values might need to be coordinated with aggregators. If partitions are used by aggregators, the amount of state the aggregator tracks is increased by the number of partitions. This represents an increase in total storage, in exchange for reducing the scope over which that storage needs to be consistent.

An aggregator could constrain the values that are accepted for this extension, rejecting reports that lack the extension or have disallowed values.

5. Privacy Budget Consumption

The gathering of reports can be modeled as the expenditure of privacy budget by a client. That is, clients treat the creation of a report from private information as a limited release of information.

Total privacy loss in this case is determined by the combination of two factors:

If aggregation includes the application of an appropriate differential privacy mechanism (that is, added noise; see [DWORK], [DAP-DP], and Section 7.5 of [DAP]), the client might rely on an understanding of that mechanism to model privacy loss. However, such privacy loss might be based on an assumption that the client contributes just one report. A complete model needs to consider the contributions of multiple reports.

A privacy budgeting system provides additional flexibility. Privacy loss associated with any task (or information release) can be adjusted to control the amount of noise that is added. A budget might be specified in terms of a metric (like the epsilon parameter in (ε, δ)-differential privacy) that is expended with each information release.

In one version of that model, a client is responsible for the management of any privacy budget. Each report represents a logical information release, contingent on being sent for aggregation by aggregators that are trusted to apply an appropriate differential privacy mechanism with the appropriate level of noise.

5.1. Privacy Budget Report Extension Format

The privacy_budget report extension (codepoint 0xTBD) encodes the amount of privacy budget that the client considers to be expended as a result of producing a report.

The value of the codepoint is an encoding of the number of milli-epsilons of budget that are expended, using as many bytes as needed to encode the value in network byte order. Each unit is a one-thousandth of an epsilon (ε) as used in (ε, δ)-differential privacy.

5.2. Privacy Budget Usage

An aggregator that is configured to apply a differential privacy mechanism can operate in one of two modes: either one where the privacy budget value is validated and reports that contain a small value are rejected; or, where the minimum privacy budget value is used to determine the parameters for the differential privacy mechanism.

In the first mode, aggregators each validate this parameter as part of validating each report. The value in the report is compared with the value configured for the task. A report that contains a value that is lower than the value configured for the task is the result of a client that expects that the aggregators will add more noise than the task configuration presently allows. Aggregators MUST reject reports with a privacy budget value that is smaller than their configured privacy budget.

Alternatively, aggregators could adjust the parameters of the differential privacy mechanism they use to match the smallest privacy budget that was included in reports. For long-running tasks that produce multiple outputs over time, it is only necessary to ensure that each output contain noise that is based on the minimum budget expenditure of the reports that are included in that aggregate.

This report extension can be used to protect reports that are conveyed from client by untrusted entities, especially where those entities might be able to choose any task, as enabled by the late_binding report extension (Section 3). This parameter ensures that the entity cannot direct reports to a task that has an inadequate differential privacy mechanism.

6. Security Considerations

The security considerations relevant to each extension is enumerated in the respective sections: Section 3, Section 4.1, Section 4.2, and Section 5.

Use of DAP is subject to the security considerations of DAP (Section 7 of [DAP]) and the VDAF that is in use (Section 9 of [VDAF].

7. IANA Considerations

Registrations for the defined report extensions need to be made, but this depends on the resolution of the TODO in Section 8.2.2 of [DAP].

8. References

8.1. Normative References

[DAP]
Geoghegan, T., Patton, C., Pitman, B., Rescorla, E., and C. A. Wood, "Distributed Aggregation Protocol for Privacy Preserving Measurement", Work in Progress, Internet-Draft, draft-ietf-ppm-dap-12, , <https://datatracker.ietf.org/doc/html/draft-ietf-ppm-dap-12>.
[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/rfc/rfc2119>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/rfc/rfc8174>.

8.2. Informative References

[ASCII]
Cerf, V., "ASCII format for network interchange", STD 80, RFC 20, DOI 10.17487/RFC0020, , <https://www.rfc-editor.org/rfc/rfc20>.
[DAP-DP]
Chen, J., McMillan, A., Patton, C., Talwar, K., and S. Wang, "Differential Privacy Mechanisms for DAP", Work in Progress, Internet-Draft, draft-wang-ppm-differential-privacy-00, , <https://datatracker.ietf.org/doc/html/draft-wang-ppm-differential-privacy-00>.
[DWORK]
Dwork, C. and A. Roth, "The Algorithmic Foundations of Differential Privacy", Now Publishers, Foundations and Trends® in Theoretical Computer Science vol. 9, no. 3-4, pp. 211-407, DOI 10.1561/0400000042, , <https://doi.org/10.1561/0400000042>.
[ORIGIN]
Barth, A., "The Web Origin Concept", RFC 6454, DOI 10.17487/RFC6454, , <https://www.rfc-editor.org/rfc/rfc6454>.
[SHA2]
Eastlake 3rd, D. and T. Hansen, "US Secure Hash Algorithms (SHA and SHA-based HMAC and HKDF)", RFC 6234, DOI 10.17487/RFC6234, , <https://www.rfc-editor.org/rfc/rfc6234>.
[SITE]
WHATWG, "HTML - Living Standard", , <https://html.spec.whatwg.org/#site>.
[TASKPROV]
Wang, S. and C. Patton, "Task Binding and In-Band Provisioning for DAP", Work in Progress, Internet-Draft, draft-ietf-ppm-dap-taskprov-00, , <https://datatracker.ietf.org/doc/html/draft-ietf-ppm-dap-taskprov-00>.
[VDAF]
Barnes, R., Cook, D., Patton, C., and P. Schoppmann, "Verifiable Distributed Aggregation Functions", Work in Progress, Internet-Draft, draft-irtf-cfrg-vdaf-12, , <https://datatracker.ietf.org/doc/html/draft-irtf-cfrg-vdaf-12>.

Acknowledgments

Roxana Geambesu noted that a binding to requester identity (Section 4.1) was an important component of a robust differential privacy system design.

Author's Address

Martin Thomson
Mozilla