Skip to main content
For step-by-step setup instructions, see Run an Arc Node.

​
System requirements

See Reth system requirements for more detail on Execution Layer configuration.
During sustained high load, such as startup or extended sync when the node is far behind, Execution Layer memory usage surges on some hardware. Systems that meet the listed requirements handle these surges without intervention. If you observe memory growth, enable backpressure to throttle execution to the speed of disk writes.

​
Snapshot download size

An Arc node bootstraps from a snapshot; syncing from genesis is not supported. Testnet snapshots are approximately 68 GB compressed for the Execution Layer and 16 GB compressed for the Consensus Layer, extracting to roughly 103 GB and 36 GB respectively. On a stable 100 Mbps connection the download takes 10–15 minutes; slower or metered connections can take hours.

​
Required binaries

Every Arc node ships these three components. Install them with arcup (pre-built binaries), build them from the arc-node source, or run them in Docker. See Run an Arc node for the full walkthrough.

​
Versions

​
Network endpoints

Your Consensus Layer connects to relay endpoints to fetch blocks from the network. Specify multiple endpoints for redundancy. For developer RPC endpoints (connecting to Arc as an application, not running a node), see Connect to Arc.

​
Exposed ports

The Execution Layer and Consensus Layer communicate through either IPC sockets or RPC. Choose one mode; they are mutually exclusive. IPC mode (default): Both processes run on the same host. Lower latency, no authentication required. RPC mode: Processes run on separate hosts. Uses HTTP on ports 8545 and 8551 with JWT authentication. Requires generating a shared JWT secret. See Run an Arc node: Run on separate hosts for configuration details. If your node is reachable from the public internet, start the Execution Layer with --public-api. This flag hides pending-transaction RPCs (a potential MEV vector) and warns if --http.api or --ws.api exposes namespaces beyond the safe set (eth, net, web3, rpc).

​
Pruning

Both the Execution Layer and Consensus Layer accept the --full flag to enable pruning. When bootstrapping from a pruned snapshot, --full is required on the first Execution Layer start to reconcile internal database tables that would otherwise fail a consistency check. After the initial startup completes, you may restart without --full if you prefer not to prune. EL pruning can increase memory usage on constrained machines. If you observe memory pressure, lower the backpressure threshold.