1 of 13

THE ORA DATA PRESERVATION SERVICE�A LIGHTWEIGHT, OPEN-SOURCE, DIGITAL REPOSITORY SOLUTION

Tom Wrobel�Oxford Research Archive (ORA), Bodleian Libraries, University of Oxford

Open Repositories 2024�Developer Track Session 3, 2024-06-06

2 of 13

ORA�(OXFORD RESEARCH ARCHIVE)

ORA is the institutional repository for the University of Oxford, founded in 2006

300,000 digital research objects incl. journal articles, theses, small-medium size datasets, and working papers

Built from Samvera Hyrax 2.5, using Fedora 4.7 for storage

https://ora.ox.ac.uk

3 of 13

ORA DATA PRESERVATION SERVICE (DPS) REQUIREMENTS

Independent of operational repository

Not dependent on single supplier/software

Co-locate object metadata and binary files

Store versioned digital objects

Data storage in an application agnostic format

Portable and re-usable

Fully de-coupled independent data store of versioned digital objects comprising both (serialised) metadata and binary files
Horizontally scalable for > 1,000,000 objects
Hold binary files up to 256Gb
To allow use by other repository services: store content in an ORA agnostic way, and use a non-ORA/Hyrax specific API
In addition, for long-term data security and availability:

Be re-buildable from its own file system (no software dependencies)
Provide access to all files for digital preservation activities
Preserve each object and object version within a single directory structure
Maintain a clear relationship between DPS files and their operational counterparts for debugging and system checks
Ensure only updates to digital objects between versions needed to be preserved (reduces cost and complexity)
Backed up using standard filesystem management tools
Interaction between repositories and the DPS using HTTPS only

4 of 13

THE ORA DPS IMPLEMENTATION

THE ORA DPS USES A FEDORA6 (V6.5) REPOSITORY WITH ARCHIVAL GROUPS AS AN API LAYER TO MANAGE AN OCFL FILE SYSTEM

THE DPS IS THE OCFL FILE SYSTEM ON DISC, NOT THE FEDORA6 LAYER

ALL CODE RELEASED UNDER MIT LICENSE, INCLUDING FEDORA6 TESTING CODE

5 of 13

ORA REVIEW AND ORA DPS SERVICES

ORA4-REVIEW-FEDORA

Fedora Repository Server

ORA4-REVIEW�Hyrax application server

ORA-DPS

Fedora6 application� server

ORA OCFL � Storage root

Digital Preservation �Microservices agents

HTTPS-only connection

Internal/system level connection

6 of 13

ORA DPS TIMELINE

Need identified

Jan 2020

Work package agreed

Apr 2023

Product brief signed off

Aug 2023

Fedora6 component ready

Sep 2023

All content in DPS

Jan 2024

Release to BAU

Jan 2024

4 months

6 months

Fedora6 performance testing and configuration

Contractor time - including performance testing

18 months

Product scoping and requirements gathering

Soft launch

New objects and edits preserved

Dec 2023

Full end-to-end code solution

Full integration with Hyrax systems�Oct 2023

Proof of concept code

First content saved in DPS�Oct 2023

Acceptance testing

Nov 2023

* The 90/10 rule applies here. We spent three months load testing Fedora6 to ensure that it would meet our needs when we deployed it, and optimising settings and practices for that initial ingest. All tests passed with flying colours. We also spent a lot of time making HTTP body_stream work. Had we known what we were trying to do, that would have been much faster.

Metadata crosswalk to serialise ORA object metadata in a single file: OraDataModel::export. 2 weeks development time
Operational code to save an ORA object in Fedora6 (create/update object, save metadata files, create/add binary files): ORA::DPS. 2 weeks development time*
Ruby on Rails gem to interact with the Fedora6 API (released under MIT licence): 4 weeks development time
Fedora6 deployment and configuration: 2 weeks development time*

7 of 13

8 of 13

SERVICE BENEFITS

centralised, content agnostic , traceable�

The ORA DPS exposes the content within to external digital preservation and system monitoring tools, such as virus checking, while retaining object and file structures. Increased backup coverage and performance.

beyond initial requirements

digital preservation

metadata and binary content�

Metadata is a first-class citizen in digital preservation, enhancing data security. Point in time versions can be downloaded

versioned digital objects

future-proofing built in�

�The application-agnostic nature of data stored in the ORA DPS means that whatever the next generation of ORA will be, this will be its DPS, and we can populate any new system from the objects in the DPS without the need to migrate

migration-proof data

9 of 13

POPULATING THE ORA DPS

ORA REPOSITORY IN DEC 2023

OBJECTS�295,000

OBJECTS W. BINARY FILES�120,000

DISTINCT BINARY FILES�250,000

REPOSITORY SIZE ON DISK�2.6TB

12 hours, in three batches

1,650,000 distinct files in the filesystem, including Fedora metadata files

832,000 ORA metadata and binary files

250,000 ORA binary files

2,000,000 �API calls

2.6TB of data

10 of 13

RUNNING THE ORA DPS�EACH MONTH

ORA REPOSITORY IN MAY 2024

OBJECTS�300,000

OBJECTS W. BINARY FILES�125,000

DISTINCT BINARY FILES�280,000

REPOSITORY SIZE ON DISK�2.7TB

3,500 object pushes

1,200 new objects

17,000 API calls per month

40GB of data

11 of 13

EC2 instance: $1000/month, API calls $10,000, S3 storage: $63,

SPECIFICATION AND COSTS

RHEL 7, 12 CPUs, 32GB RAM, 8TB Dell Compellent Storage

build

cost

aws migration

aws monthly

On premises hosted with no stated costs

EC2: $250, API: $85, S3 Storage: $63

12 of 13

WITH THANKS TO

CottageLabs, providers of many of the components listed here, especially Anusha Ranganathan
The ORA Development Team
The Bodleian Digital Systems and Services Infrastructure Team
The Fedora6 community

13 of 13

GET IN TOUCH

thomas.wrobel@bodleian.ox.ac.uk

MORE INFORMATION, CODE, AND LINKS TO RESOURCES

https://github.com/tomwrobel/OR2024