IPFS Pinning Service on an Anubiz Romania VPS

The default Kubo datastore on flatfs becomes a problem fast: every block is a separate file, inode pressure on ext4 climbs quickly, and you hit the practical wall around 4-6 TB pinned. For anything serious switch to the badger v4 datastore, which keys blocks into LSM-tree SSTables and survives at the tens of terabytes level with predictable compaction.

Plan storage in three layers. Hot blocks (recently fetched, frequently served) want NVMe with 50k+ random read IOPS - the Anubiz Romania VPS V tier (NVMe, $99.99/mo) covers this. Warm blocks (pinned by paying customers but rarely fetched) can sit on a second SATA SSD volume that we attach on request. Cold blocks (archive pins, Filecoin deal staging) belong on spinning rust at a different price tier - reach out before you scale past 8 TB.

Estimate disk growth at 1.4x raw pin volume to account for badger LSM amplification, refs index, and the announce queue. A 2 TB plan handles roughly 1.4 TB of real content before you should consider an upgrade.

The painful part of running a pinning service is not pinning - it is serving. A popular CID can fan out to dozens of peers in seconds and saturate a small egress pipe. Romania transit pricing lets us ship 32 TB monthly on the VPS V tier without metering. If you need genuinely unlimited, the dedicated server line is the right escape hatch.

Tune Bitswap: bump MaxOutstandingBytesPerPeer, raise HighWater on the connection manager to 900 for a node that announces tens of thousands of CIDs, and enable accelerated DHT client. Disable the public gateway on the same port - run gateway traffic behind nginx with rate-limit zones, otherwise a curl loop from a single ASN can wedge the node.

Set ConnMgr.Type=basic only on tiny nodes. Production pinning uses type=adaptive with grace period 30s and high water 1200.

One pinning node is a single point of failure. The IPFS Cluster project gives you Raft consensus across two or three nodes that share the pin set, so a reboot or disk swap on one box does not interrupt availability for paying customers. Run the cluster service on the same VPS as your Kubo daemon and front it with a small HTTPS API for your billing layer.

For the second cluster member we recommend a different jurisdiction - Iceland, Netherlands, or Finland - to spread legal risk. The cluster gossip protocol works fine over WAN as long as you keep the proxy port reachable. Use the cluster secret as a shared key; never expose the REST API to the public internet.

Pinning service operators are intermediaries under Romanian implementation of the EU Digital Services Act. That means you receive a NTD (Notice and Takedown) request, you act in a reasonable window, and you keep records of action taken. Anubiz Host does not see what your customers pin and will not unpin CIDs at our level - the request flows to you, the operator, and you handle it. We pass on court orders that name our IP space.

Be honest with customers about what we cannot promise. The infrastructure is privacy-respecting, not law-immune. CSAM is removed on first credible report, no exceptions, no appeal. Build that into your terms of service before you take the first customer.

Order an Anubiz Romania VPS V or larger with crypto (BTC, XMR, LN) or fiat. Activation is 12-20 minutes. We ship Debian 12 with Docker Compose ready; the Kubo + cluster stack is a 10-minute follow-up. Migrating from an existing Pinata or web3.storage setup? Use ipfs-cluster-ctl pin add --rmin 2 --rmax 3 against the new cluster after exporting your pin list as JSON.

Related reading: Romania VPS product page, offshore VPS lineup, anonymous hosting overview, dedicated server tier for >10 TB pin sets.