IOSP 2026 · Leiden
Resilient Data
At-risk research datasets, archived on hardware researchers own and run themselves. What this page shows today is the pre-workshop network, a small demonstration we stood up ahead of IOSP 2026; at the workshop in Leiden this October, up to 20 researchers stand up their own nodes and grow it into the real thing. Everything below is the network’s live health readout, reported by the clusters’ own gossip.
Want a node of yours in this readout? Sign up for the workshop.
iosp-nodes
Raspberry Pi nodes. Today the founding dry-run Pi; after the workshop, the Pis researchers take home.
Datasets · 5
sampled 19:10 UTCiosp-laptops
Researchers’ own laptops. Today a single facilitator laptop; participants’ laptops join at the workshop.
No member heartbeats in the last 24 hours.
Datasets · 4
sampled 19:10 UTCRecorded by node-00 (observer) — heard 100% of passes. The observer’s stored copies are not counted as replicas of this cluster.
What this instrument can see: the observer samples cluster gossip every minute; a node counts as heard if its heartbeat was live at that moment. Dips shorter than the sampling gap can pass unseen. A dashed divider marks a period when the recorder itself wasn’t sampling — absence of evidence, not evidence of absence. “Free” is each node’s configured archive budget remaining, not its whole disk. Strips show the most recent 180 passes; uptime percentages cover the window of passes the readout holds. Dataset rows come from each cluster’s own pin tracker, sampled every ~5 minutes: “serving” counts members whose node reports the item pinned, and the observer’s copies are not counted in iosp-laptops. A “✓ served” mark is stronger than the count: every 30 minutes the observer picks members and makes them actually deliver a block of the dataset over the wire — proof of serving, not a claim. The observer never probes its own copies, so marks appear only where another member answered.