Bring-up weave for the single-particle cryoEM pre-processing demo. It provisions a multi-GPU FABRIC VM with the CryoEM Without Borders storage volume, plus a receiver VM connected over a dedicated L2 link.

What the weave does

1. Picks a GPU host with NUM_GPUS free GPUs of one model, using live per-host availability (preference: A40 > A30 > RTX6000 > T4).
2. Creates the GPU VM and attaches the CryoEM Without Borders CephFS project volume (the Globus collection, mounted under /mnt/cephfs/west/…_cryoem-without-borders).
3. Installs the software stack:
   • NVIDIA driver and CUDA
   • MotionCor3 and RELION 5.0 (built from source)
   • CTFFIND 4.1.14
   • A Python preview environment
4. Adds a CPU receiver VM at RECEIVER_SITE (FIU; a stand-in to be replaced by CAPE), connected by a dedicated site-to-site L2 link (XFER): private subnet, TCP-tuned, jumbo frames when the path allows, with SSH trust from the GPU node.
5. Extends the lease, writes results/slice_info.json, and exits. It waits once for the slice to reach StableOK and does not poll after that.

The experiment

The experiment itself lives in cryoem_experiment.ipynb:

• Frame alignment with MotionCor3 on all GPUs
• CTF estimation with CTFFIND4
• Before/after comparison figures
• iperf3 baseline, then rsync transfer of the results to the receiver's local disk with checksum verification
• Optional RELION pipeline

Notes

• With KEEP_SLICE=true (the default), neither Stop nor the cleanup script deletes the slice.