Why does Quip Network run its testnet on Flux instead of AWS?

Quip Network runs the public side of its Aglais testnet on Flux, a decentralized computing network, for about $55 a month. That covers seven services: the bootnodes, an EVM endpoint, the faucet, two block explorers, the web portal and a node that publishes chain snapshots. Renting the same servers from AWS would cost about $1,060 a month.

Price is the first reason we chose Flux. The others are what it does for small services like ours: it handles certificates for our web front ends, lets us choose where services run, and doesn't bill traffic. The tradeoff is that we build every service to recover on its own when it starts on a new machine.

What the testnet needs

Aglais is the public testnet of Quip Network. Developers connect wallets to it, request test tokens and follow blocks in its explorers, so its public services have to be reachable from anywhere. Seven services do that work:

• Bootnodes: three validator nodes and a standby that can take over a validator slot, each with 4 vCPU, 8 GB of memory and 150 GB of disk.
• EVM endpoint: a non-validating node that answers Ethereum-style JSON-RPC requests.
• Faucet: hands out test tokens.
• Explorers: Blockscout for the EVM side of the chain and Subscan for the Substrate side.
• Web portal: our fork of the Polkadot.js Apps interface, with a post-quantum signer.
• Snapshot publisher: a node that publishes a fresh copy of the chain database every 6 hours, so new nodes can start from it.

Together they use about 33 vCPU, 76 GB of memory and 1.1 TB of disk.

What the same servers cost elsewhere

We priced the same vCPU, memory and disk on AWS and DigitalOcean, taking the lowest-priced on-demand option that fits each service, with a public IPv4 address where the service takes direct connections.

The figures cover servers only and leave out backups and object storage. They compare resources only: uptime guarantees, support and performance differ between providers and aren't priced here. Flux states that "prices are in US dollars per month", and paying in FLUX gets a 5% discount.

We saw the same gap when we moved two web backends that sit outside the testnet from DigitalOcean to Flux. Their droplets listed at $36 a month together. On Flux the pair costs about $5 a month.

Why Flux fits our testnet

Flux is a decentralized cloud: its servers, called FluxNodes, are run by independent operators rather than housed in one company’s data centers. Operators stake FLUX tokens as collateral and earn block rewards for keeping their nodes online, and Flux’s docs report more than 10,000 nodes in 66 countries, run by over 560 independent providers.

Developers rent that capacity through FluxCloud, which deploys containerized apps across the network and shifts a workload to another node when one fails.

Five things make Flux a good fit for services like ours.

1. Certificates for web front ends are handled for us. Flux's Orbit page says "every plan also includes custom domains with managed TLS", and that "SSL is handled automatically through the Flux reverse proxy network". Our faucet, explorers and web portal hold no certificates of their own.
2. We choose where services run. Flux offers a static IP option for apps, which suits the bootnodes and the EVM endpoint that other machines connect to directly. Its geolocation rules let us allow or forbid continents, countries or regions, which is how we keep our monitoring probes in North America, Europe and Asia.
3. Traffic isn't billed. The FluxCloud introduction lists "no ingress or egress fees". A public RPC endpoint and two block explorers send data all day, and AWS bills outbound data by the gigabyte.
4. The web portal deploys from git. Flux describes Orbit, its git-based deployment service, this way: "Push your code and Orbit automatically builds, deploys, and scales your app". Our web portal is built and served straight from its git branch.
5. No single hosting company owns the machines. Flux says its network "is highly decentralized by design", with nodes operated by "independent infrastructure providers". For the testnet of a decentralized network, that matters to us.

What it costs besides the bill

Flux restarts an app on another node when its node fails. Its docs say "if one FluxNode fails, another will automatically take over". We plan for the hardest version of that event, where a service starts on a fresh machine without its old data.

Each service has its own way back. Chain nodes download the newest chain snapshot from object storage and sync the rest from peers. The explorers and the faucet rebuild their state from the chain, and Orbit rebuilds the web portal from git. A move still means downtime while a service restores itself, often tens of minutes, and most of our services run as a single instance with nothing to fail over to.

The bootnodes and the EVM endpoint take direct connections, so they hold their own TLS certificates. They keep those certificates in encrypted Tarsnap archives, restore them when they start and renew them in the background, so a restart doesn't wait on a certificate authority. We made that a rule after three of our bootnodes restarted at once and their certificate requests ran into issuance limits.

We also wrote that recovery tooling ourselves and keep it working, along with a small coordination service that assigns the three validator slots to running bootnode instances. On AWS or DigitalOcean, a storage volume that outlives the machine would have done part of that work for us.

Final thoughts

Flux runs the public side of our testnet for about a twentieth of what AWS would charge for the same servers, and about a ninth of DigitalOcean's price. That saving comes from services built to start over on a new machine, and we count the engineering behind them as part of the cost. For small services that can rebuild their own state, the trade has worked for us.

Try the Aglais testnet today.