1 | 1.1.1 | Minimum Viable Product (MVP) for registering metadata in the repository and connecting the metadata to the data in the research computing remote storage (NESE), including Globus endpoints | 15

[sync-by-unito]

References:

• Backlog Grooming Board

Problem Statement

Prior to this work, Dataverse is capable of storing up to about 1TB in S3.

Proposed Solution

The first part was to integrate Globus as a large file transfer mechanism, into Dataverse. This was done by building on the work already done in the Borealis, the Canadian Dataverse Repository (Formerly Scholar’s Portal Fork of Dataverse) to allow Dataverse to integrate with Globus. This integration allows files bigger than a terabyte to be transferred from within Dataverse and to an S3 store via globus.

The second part addresses very large files; items up upward of a petabyte or so are not realistic for DV to store. This solution enables Dataverse to manage datasets where one or more of the files is referenced rather than being directly stored within a Dataverse repository. In this solution, the large file remains in its original location and then is referenced from Dataverse.

Acceptance Criteria

• Discussion with Dataverse community members with related work,
• Set up Globus environment at NESE
• Design and implement code to call API to interact with Globus endpoints,
• Test integration

Associated Issues:

• Release Notes related to HDC 1 and 3A/3B dataverse#8894
• GDCC/Globus and Big Data Support dataverse#8891
• Globus: DatasetVersion in dataset level externaltool dataverse#7740
• Globus: Identify and add configurable button(s) for Upload and Download actions on Dataset page dataverse#7626
• Large file download support dataverse#7524
• Simple support for remote web stores dataverse#7325
• External data support: Globus endpoints (deposit and indexing) dataverse#5994
• Enhance Globus support for remote endpoints and tape stores dataverse#9123
• Weird property key modification in Bundle.properties file v5.12 dataverse#9189

See comments below for latest update.

---

┆Issue is synchronized with this Smartsheet row by Unito

[mreekie]

This issue represents a deliverable funded by the NIH
This deliverable supports the NIH Initiative to Improve Access to NIH-funded Data

Aim 1: Support the sharing of very large datasets (>TBs) by integrating the metadata in the repository with the data in the research computing storage

An increasing number of research studies deal with very large datasets (>TB to PBs). When the study is completed or ready to be distributed, it is not always feasible nor desirable to deposit the data in the repository. Instead, in this project we propose to publish the metadata to the repository for discoverability of the study and access the data remotely from the research computing cluster or cloud storage. In this scenario, the data does not need to be downloaded to the user’s computer but can be viewed, explored, and analyzed directly in the research computing environment. The Harvard Dataverse repository will leverage the Northeast Storage Exchange (NESE) and the New England Research Cloud (NERC) to provide storage and compute for these very large datasets by finding and accessing them through the repository and keeping the metadata and data connected via a persistent link. These two services - NESE and NERC - are large-scale multi-institutional infrastructure components of the Massachusetts Green High Performance Computing Center (MGHPCC) -- a five member public-private partnership between Boston University, Harvard University, Massachusetts Institute of Technology, Northeastern University, and the University of Massachusetts. MGHPCC is a $90 million facility that has the capacity to grow up to 768 rack, 15 MW of power and 1 terabit of network capacity in the current 95,000 sq. ft data center. One of the key integration points to support large data transfers is to incorporate Globus endpoints. Globus is a distributed data transfer technology developed at University of Chicago that is becoming ubiquitous for research computing services. This will allow the realistic transfer of TBs of data in less than an hour. Globus will also be a front end of NESE Tape, a 100+ PB tape library within MGHPCC. The integration of the repository with research computing is one of the components of a Data Commons that will facilitate collaboration, dissemination, preservation and validation of data-centric research.

Related Deliverables:
2.1.1 | 1 | Test and apply metadata registry for large datasets and integration with research computing for a few NIH-funded projects, piloting the cost recovery model | 10
2.1.1 | 1 | Test and apply metadata registry for large datasets and integration with research computing for a few NIH-funded projects, piloting the cost recovery model | 10
4.1.1 | 1 | Assess cost recovery model | 10
4.1.2 | 1 | Improve metadata registry UX/UI and integration with remote storage and computation based on user feedback from the previous years | 10

---

This work also represents a deliverable funded internally.

Harvard Data Commons MVP: Objective 1
Objective: Integrate Harvard Research Computing environments and Harvard repositories to facilitate publishing data and/or metadata throughout the research project lifecycle
Publish datasets, w/ data and metadata (GBs size)
Publish metadata only and reference data in RC (TBs size)

• Work package 1: Review and assess an existing open-source Globus connector tool
• Work package 2: Implement the rest of the connector tool to support use cases A and B
• Work package 3: Extend Dataverse UI to support the connector tool in a user-friendly way
• Work package 4: Beta test with real data and users

This picture shows how the the Harvard Data Commons work maps to Dataverse work.

This is a closer look at the Harvard Datacommons work: GDCC DataCommons Objective 1 Task Tracking

• This is not a public link

[mreekie]

Next step:

• Sync up with @qqmyers to summarize this work in a non-technical way.
• Are we finished with this NIH deliverable? (yes)
• write a final update and add it to the description.
  • Update the notes from discussion/update from Jim
  • Run them by Jim & Phil for correctness.
  • Incorporate changes

[mreekie]

Summary
This work enabled a minimum viable process for supporting much larger files.
The code associated with this minimum viable process is included in Dataverse 5.12.

Prior to this work, Dataverse was capable of storing up to about 1TB in S3.

The first part was to integrate Globus as a large file transfer mechanism, into Dataverse. This was done by building on the work already done in the Borealis, the Canadian Dataverse Repository (Formerly Scholar’s Portal Fork of Dataverse) to allow Dataverse to integrate with Globus. This integration allows files bigger than a terabyte to be transferred from within Dataverse and to an S3 store via globus.

The second part addresses very large files; items up upward of a petabyte or so are not realistic for DV to store. This solution enables Dataverse to manage datasets where one or more of the files is referenced rather than being directly stored within a Dataverse repository. In this solution, the large file remains in its original location and then is referenced from Dataverse.

What's left to complete the overall minimum viable process?

• The Harvard Globus Endpoint is at the beginnings of it's setup in NESE

---

What's not in this deliverable?

• The software has not seen use with users. The Borealis, Dataverse Globus UI, was created as a proof of concept. There were some modifications to enable this work to be completed, but the interface is still at the level of a proof-of-concept from the point of view of usability.
• The under-the-hood setup for using Globus with a Dataverse instance is still complicated.

[mreekie]

Who:

• Jim
• definitely Len
• There's discussion here about what's left and how to handle that.
  • e.g. Maintenance in NESE\
• Leonid

[mreekie]

Updated today. Met with Stefano and Len.

• There was a second meeting with Scott this week.
• The establishment of the Globus endpoint at NESE is one of the last big things to be done for this deliverable.

(1.1.1) Now that these changes have been released in Dataverse 5.12, the focus has shifted to setting up a production environment for researchers to use the remote large data support. Discussions are continuing over the next weeks around allocating storage and establishing a Globus endpoint for Dataverse from the Northeast Storage Exchange (NESE).

[mreekie]

Globus support needs to be revised.
You can upload the data.
If you get the data, it takes so long that you get a timeout.
This use case needs more work.
There are discussions going on with the library about who will do the additional work and who will pay.
There are discussions around payment for the storage iteself

[qqmyers]

Slow download is not a problem with the current implementation using a Globus S3 connector over a file system. It will be a limit when the underlying storage is a tape system, where using the Globus S3 connector may also not be as useful as using their file connector. See IQSS/dataverse#9123 for details. scholarsportal/dataverse-globus#2 is also relevant.

[mreekie]

November: 2022

(1.1.1) Focused shifted to setting up a production environment for researchers to use the remote large data support. Discussions are continuing over the next weeks around allocating storage and establishing a Globus endpoint for Dataverse at the Northeast Storage Exchange (NESE).

[mreekie]

Discussions are proceeding around next steps

Summary from talking with Jim,

There was a meeting Monday where Scott Yokel, Len, Stefano, Jim and others went over the proposal/requirements/design options doc Jim put together. Planning to talk with Scholars Portal/Borealis about their interest in this as well.

---

Next step:

• Include a brief update.
• Follow-up with Stefano on this deliverable and any follow-on work in the second year.

[mreekie]

Reviewed today. Closing out 2022. This work closes out the end of February.

• Discussion with Dataverse community members with related work,
• Set up Globus environment at NESE

The following two deliverables are still being worked.

• Design and implement code to call API to interact with Globus endpoints,
• Test integration

We have an MVP representing a connection from HDV and globus but it does not support large files yet. This requires additional design and implementation steps.

[mreekie]

Last updated: Thu Dec 15 2022 before I left for the holiday
Report: Dec 2022

Planning continues around supporting the Globus endpoint for Dataverse at the Northeast Storage Exchange (NESE) and moving beyond the MVP. The MVP enables connection from Harvard Dataverse to the storage but does not support large files.

90%

[mreekie]

priority discussion with Stefano;

• Can Jim help us size the last two issues for this?