EPSCI

Group Meeting�

October 20, 2026

1

EIC Computing and Software

2

Accomplishments

1. one …
2. two …
3. three …

Validation

• Completed ePIC computing review last week. Did not receive any homework directly related to the validation efforts (hooray).

AI4EIC

• Convening tracks
• Initial meetings had/set

EPSCI Group Meeting ,October 20, 2026

Hydra

3

Accomplishments

1. one …
2. two …
3. three …

Outreach

• Jens’ talk at ORNL mentioned Hydra
• SNS reached out through Jordan looking into Hydra

Development

• First vision transformer model is trained on PS_occupancy plots
• Frontend: End-to-end (E2E) and integration testing for major frontend workflows is in progress.
• Goal: Establish sufficient test coverage to catch frontend regressions during future agent-driven development.
• Next steps:
• Complete E2E and integration tests for all pages, in priority order.
• Implement the Admin page.
• Optional: Develop an agent-driven workflow from issue to implementation, with human approval after testing. Existing regression coverage would allow reviewers to focus on the new behavior and code quality.

EPSCI Group Meeting ,October 20, 2026

Coupling Experiment to Accelerator Controls

4

Accomplishments

1. one …
2. two …
3. three …

GlueX

• Previous meetings were canceled due to conflicts (ePIC computing review, Hackathon at ANL)

MOLLER

• Additional meetings held to formalize dataset descriptions and to reduce the number of EPICS related features (over 1k were pulled from the archive)
• Meeting today with He Zhang to quantify sim2real gap between initial beam conditions and ELEGANT model simulations
• If tolerable, we will move forward with developing an environment for use in SOCT based on ELEGANT model
• Else, come up with plan to address sim2real gap and hopefully reduce

EPSCI Group Meeting ,October 20, 2026

DAQ: Streaming Readout + CODA

5

Hall B SRO

1. Working again after replacing
2. All 18 crates reporting over ESnet LB
3. HVs reportedly on in ECAL(?)
4. FT3-FT5 can read noise data
5. Not ready for waveform data yet

DAQ Lab Test Setup

• Finishing environment setup on indra-s1 (CODA 3.11_beta) - today
• Getting VTP streaming DAQ configs running
• Reproduce Dave’s setup
• Stream cosmic data (2 paddles) + signal generator signals to Perlmutter
• Account request approval in process

Container / Data Pipelines

• New container for frame building, receiver, data files, tests…
• Loopback tests: indra-s2, perlmutter, ebpf2201, ejfat-fs
• Using ESnet LB: indra-s2, perlmutter, ebpf2201 (NOT ejfat-fs)
• Can transfer mock output of vtps (raw evio-4 frames),�Or mock output of frame builder (evio-6 aggregated frames)
• Can receive, but can’t reassemble from DAQ yet though 🙁
• (data files for simulated senders included)

​

• Receiver w/ old E2SAR on indra-s2 still works
• 100% packet transmission for 1-2 crates, few seconds
• All 18 crates seen by receiver
• Not building with all 18 yet, debugging

EPSCI Group Meeting ,October 20, 2026

6

DAOS and DAQ

E2SAR+DAOS. Steps and experiments on ebpf2201 → (LB) → ebpf2203.�https://github.com/JeffersonLab/e2sar-daos

1. Install e2sar (0.4.0rc1) released package on both sender & receiver nodes
2. Fio test with DFS engine on ebpf2203: 22.2 GB/s, 178 Gbps [ebpf2303 → daosfs2302, 200 Gbps IB].

fio_dfs_2302_write_bs1M_nj4_iod64_nf1_20261015_1340.json

• Change default TCP/IP mem buffer size to 1 GiB, on both sender (ebpf2201) and receiver (ebpf2203).�sudo sysctl -w net.core.rmem_default=1073741824 net.core.wmem_default=1073741824
• iperf3 N2N for 4 streams with UDP and no packet loss, ~20 Gbps [ebpf2201 → ebpf2203, 100 Gbps ETH]
• e2sar_ft N2N: no event loss with --smooth flag, 10–12 Gbps [ebpf2201 → ebpf2203, 100 Gbps ETH]
• e2sar_ft sender (sending a 1.78 GB evio file), daos receiver, 10.7 Gbps
• e2sar_evio sender (sending frames, 256 frames as an Event), daos receiver:
1. Highest loseless: 33.7 Gbps through LB; 24.9 Gbps N2N
2. Loseless every run: 20 Gbps through LB
3. Highest with loss: 59.8 Gbps through LB

Oct/6 – Oct/20 Accomplishments

https://sc26.conference-program.com/session/?sess=sess129

EPSCI Group Meeting ,October 20, 2026

JANA2

7

Nathan + JANA2

1. Fixing a bug Raiqa found: Total processed event count sometimes includes timeframes, not just physics events�
• The processed event count logic predates JEventLevels
• Counts all events output from arrows flagged as “sinks”, as opposed to from ports
• Refactoring: New design produces a distinct event count + rate for each JEventLevel
• One tricky question left involving open vs closed intervals

​

• JANA2/ERSAP integration
• Rebuilding Vardan’s “ersap-jana2” vibe-coded example using the ‘dws’ LDRD project as a test case/worked example

Raiqa

• Merged all changes related to parser and translation extraction into plugins from my fork into main, resolving all merge conflicts.
• Updated all subrepositories that depend on core JCE to incorporate its latest changes, and verified them through testing and use.
• Improved JCE performance by switching from the default Debug build to `RelWithDebInfo`:
• Increased EVIO parsing rates from approximately 20k to 200k events/s.
• Achieved approximately 60k–100k events/s with translation enabled, depending on the raw data and parsing required.
• These results make JCE a stronger option for fast processing, even with a single processing thread.
• Evaluated several ROOT file-writing approaches; sequential RNTuple writing with four compression threads performed best.
• TODO:
• Reducing new detector translator addition to configuration only

SIDIS = file with multiple average size events

Moller = High frequency smaller events file

EPSCI Group Meeting ,October 20, 2026

Genesis Mission / American Science Cloud

8

Accomplishments

1. one …
2. two …
3. three …

MOAT Phase II

• Tanya submitted JLab FWPs
• Work is “supposed to” begin November 1st.

BPSW

• Successful hackathon was had!
• Insert rant about the guest house
• Leadership focused on the IPR

Raiqa - AmSc D2.2

• Updated [mya-amsc.jlab.org](http://mya-amsc.jlab.org) to use publicly available mysampler data:
• Added scripts to convert CSV files to SQL and load the data into the demo Docker database.
• Enhanced the frontend with features to:
• Delete previous jobs and retry transfers.
• Initiate transfers for query-only tasks.
• View details of past queries, including those associated with transfer-only files.
• Search inline for collections and transfer destinations.
• Scheduled an October 30 meeting with David, Laura, and Adam to finalize MYA data details before starting data card preparation.

Mike - AmSc D2.1

• Took a little time getting access to Test bed, slurm account
• Have been running basic tests to understand workflow
• Running workflow with hall D data
• Tutorials, etc…

EPSCI Group Meeting ,October 20, 2026

Miscellaneous

9

SPINDLE/LDRD

1. Meeting with MAJ tomorrow afternoon to access testbed for HPDF onboarding
2. Revising budget and milestones/success metrics due to review committee’s request. Due tomorrow COB.

GNNs for Tracking

• Data Science has a GNN training pipeline for track fitting and finding. Nadia Fomin (UTK) reached out asking if her student Jordan (we know from Hydra) could work with them on transferring this pipeline to another detector from the TDIS experiment. Initial meeting was set up with physicists on the experiment to discuss what they are actually trying to do.

MCwrapper

• Was inspired at the hackathon. New visualizations and a health dashboard�https://halldweb.jlab.org/gluex_sim/Dashboard.html

EPSCI Group Meeting ,October 20, 2026