Research
Instalación, configuración y utilización de un nodo Bitcoin en Linux
The paper addresses the practical gap between "run a Bitcoin node" as generic advice and the concrete technical decisions a node operator actually faces: which build options to enable, where to place…

- arXiv
- 2601.09748
- Published
- 2026-01-12
- Authors
- Jose Eduardo Ulloa, Diego R. Llanos
AI summary
Overview
Research area: Blockchain infrastructure and systems engineering — specifically the operation of a Bitcoin full node (Bitcoin Core) on GNU/Linux, its on-disk data structures, and the performance effects of node configuration parameters.
Technical level: Intermediate. The paper assumes familiarity with the Linux command line, compiling software from source, and basic Bitcoin concepts, but explains node/miner distinctions and file purposes from first principles.
Scope (one sentence): A hands-on, empirically documented walkthrough of compiling Bitcoin Core v29.0 from source on Ubuntu, running it as a systemd service, completing a full initial block download (IBD), auditing every file and directory the node produces, and measuring how txindex and prune change node behaviour and storage footprint.
What This Paper Is About
The paper addresses the practical gap between "run a Bitcoin node" as generic advice and the concrete technical decisions a node operator actually faces: which build options to enable, where to place the blockchain data, which configuration parameters matter, and what each one costs in disk, RAM, or query capability. The authors set out to document the full lifecycle — source compilation, service configuration, synchronization, and file-system output — and then to test configuration parameters empirically on the same running node. The stated goal is to produce a reproducible guide plus a baseline of measured resource figures that future node-performance research or blockchain-query tooling can build on.
Key Contributions
-
A reproducible source-build procedure for Bitcoin Core v29.0 on Linux, using CMake with
-DWITH_SQLITE=ON -DENABLE_WALLET=ON -DWITH_ZMQ=ON, an isolateddatadirat/opt/BLOCKCHAIN,rpcauthcredentials generated withrpcauth.py, and asystemdunit for persistence and restart-on-failure. -
A measured full IBD profile: a bash script polling
bitcoin-clievery 30 minutes, logging block height andverificationprogressto CSV, capturing a from-scratch sync from block 304263 to height 894549. -
A systematic audit of the Bitcoin Core data directory, explaining the function and observed size of
banlist.json,blocks/and its LevelDB index,chainstate/,debug.log,fee_estimates.dat,indexes/txindex/,peers.dat, andsettings.json. -
Controlled before/after experiments on configuration parameters (
txindex,prune, and — per the abstract's scope —dbcache,maxmempool,maxconnections), measured on the same node with a deliberately chosen pair of test transactions: one old (block #100000) and one recent (block #898576).
Main Findings
-
IBD completed in roughly 23 hours. Processing began 29 April at 05:54 from block 304263; at 30 April 05:00 the height was 894549, meaning approximately 590,000 blocks downloaded and validated in that window. The authors state this demonstrates a full node can sync in under 24 hours given adequate hardware, network, and configuration.
-
verificationprogressplateaus near the end of sync by design. The metric rose continuously toward 1 but stabilized or oscillated slightly near completion, which the authors attribute to Bitcoin Core's internal progress estimator recalibrating against recent network activity — not to an error. Completion required both blocks = headers andinitialblockdownload = false. -
Full (unpruned) chain storage measured at ~471 GiB. The
blocks/directory held more than 9,000blk/revfiles with mode0600, owned by thebitcoinuser, created progressively over the sync. -
UTXO set snapshot at block 847254:
txouts82454865,transactions89805073,disk_size447277290 bytes (~447 MB),bogosize6169726781,total_amount19975766.22532526. -
txindex=1costs ~58 GB but enables txid-only lookups. After switching totxindex=0and restarting,getindexinforeturned an empty object, andgetrawtransaction <txid> 1failed with error -5 ("No such mempool transaction. Use -txindex or provide a block hash…") for both the old and recent test transactions. Supplying the block hash made the same lookups succeed, confirming that disabling the index saves disk and speeds startup but removes txid-only historical queries. -
prune=550shrankblocks/to 796M and made old blocks unretrievable. After restart,getblockchaininforeturned error -28 ("Pruning blockstore …") during the initial prune, which took a few minutes of unavailability. Post-prune state reportedpruned: true,pruneheight: 898283,size_on_disk: 553096761,prune_target_size: 576716800,automatic_pruning: true. Requesting block #100000 afterwards returned "Block not available (pruned data)", while the recent block remained accessible.chainstate/occupied about 4.2 GB. -
debug.loggrows large and persistent. The file was approximately 210 MB with 909,217 lines, created 29 April 2025 and last modified 23 May 2025, indicating uninterrupted operation. Its opening lines record Bitcoin Core version v29.0 (release build), the SSE4/SSE4.1/AVX2 SHA256 implementation, RdSeed/RdRand entropy sources, thetxindex="1"config argument, and "Using at most 125 automatic connections". -
fee_estimates.dat(~250 KB) is flushed roughly hourly, with log entries of the form "Flushed fee estimates to fee_estimates.dat".estimatesmartfee 1returnedfeerate0.00003145 withblocks2;estimatesmartfee 0was rejected with error -8, "Invalid conf_target, must be between 1 and 1008". -
peers.dat(~2.6 MB) supported 10 active peer connections, inspected viagetpeerinfo, with fields includingservicesnames(NETWORK, WITNESS, NETWORK_LIMITED, P2P_V2),relaytxes,pingtime,minping,subver,bytessent/bytesrecv, andconnection_type. -
banlist.json(221 bytes) andsettings.json(~190 bytes, mode0600) remain at their auto-generated defaults, each containing only a warning against manual editing;banned_netswas empty. -
Results for
dbcache,maxmempool, andmaxconnectionsare not available in the supplied text. The content is truncated at the start of thedbcachesection (a value range beginning "approximately 450–50" is cut off), so no figures for those three parameters can be reported here.
Methodology in Plain English
The authors took a single commodity desktop — an Intel Core i5-8400 with 6 cores, 16 GB of RAM, and a 2 TB disk — and turned it into a full Bitcoin node. Instead of downloading a prebuilt binary, they cloned the official Bitcoin repository, checked out tag v29.0, installed build dependencies from Ubuntu's repositories, and compiled with CMake, explicitly enabling the wallet, SQLite, and ZeroMQ support so that the resulting node could later serve as a data source for analytics tools.
They put the blockchain on a dedicated directory (/opt/BLOCKCHAIN) owned by a dedicated bitcoin user, wrote a minimal bitcoin.conf with txindex=1 and generated RPC authentication, and registered the daemon under systemd so it starts at boot and restarts on failure. During the first sync they ran a small bash loop that asked the node for its block height and verificationprogress every 30 minutes and appended the answers to a CSV — a low-effort way to get a time series of sync performance.
After sync, they listed and measured every generated file and folder, using ls -lhR, du -sh, stat, file, and targeted bitcoin-cli calls (gettxoutsetinfo, getpeerinfo, getblockchaininfo, estimatesmartfee, getindexinfo, getrawtransaction). For the configuration experiments they changed one parameter at a time on the same node, restarting between changes and observing the effect immediately before and after. Because txindex=1 is incompatible with pruning, they ordered the tests deliberately: disable txindex first, then enable pruning, then proceed to the remaining parameters. They picked one transaction buried in block #100000 and one in block #898576 as fixed probes, so that pruning's effect on old versus recent data could be compared directly.
Why This Matters
The paper provides a concrete, measured baseline for what a home-scale Bitcoin node actually costs in time, disk, and query capability — figures that are usually asserted informally rather than recorded. For research, it demonstrates that node-level instrumentation (periodic polling of verificationprogress, file-system accounting) is cheap to implement and yields data suitable for comparing configurations or identifying bottlenecks during IBD.
Real-world applications:
- Block explorer and analytics backends: the
txindexexperiment quantifies the trade-off (about 58 GB of index) for services that need to look up arbitrary transactions bytxid. - Low-resource and embedded deployments: the
prune=550results (blocks directory reduced to 796M, node still fully validating) show what storage floor is achievable without losing consensus participation. - Node operator guidance and onboarding material: the step-by-step build,
systemdunit, and file-by-file directory explanations double as a teaching reference. - Fee-estimation and peer-connectivity diagnostics: the
fee_estimates.datandpeers.datsections show how to verify that a node's fee suggestions and P2P connections are healthy.
Industry relevance: self-hosted validation underpins custody, payment processing, and regulatory verification workflows where relying on a third-party API is unacceptable. Documenting the resource envelope and the query restrictions of each configuration helps teams size infrastructure and pick parameters deliberately rather than by default.
Future Directions
- Complete and report the
dbcache,maxmempool, andmaxconnectionsexperiments. The abstract promises these; the supplied text stops before any results, so the effect of cache size on IBD throughput, mempool memory behaviour, and connection limits remains an open measurement. - Repeat the sync under contrasting configurations (with and without
txindex, with and without pruning) and compare the CSV time series, as the authors themselves suggest, to isolate how each option shifts sync duration and resource use. - Extend the monitoring beyond a single machine and a single run. The paper's own framing calls for comparing scenarios and identifying bottlenecks; a multi-hardware, multi-network baseline would make such comparisons meaningful.
- Build the query tooling the data directory enables. With
txindexenabled and ZeroMQ notifications available in the build, the natural next step is a blockchain query or visualization layer, which the authors explicitly name as a motivation for their configuration choices.
Target Audience
Linux system administrators and technically comfortable Bitcoin users who want to run and understand their own node; students and instructors in systems, distributed-systems, or blockchain courses looking for a reproducible lab exercise; and developers or researchers building blockchain data-query tools who need to know exactly what Bitcoin Core writes to disk and what each configuration parameter costs. Readers seeking protocol theory, cryptography, or consensus-algorithm analysis will not find it here — this is an operations and measurement paper.
Authors’ abstract
This paper documents the installation, configuration, and operation of a full Bitcoin node in a Linux environment, from manual compilation of the source code to complete synchronization with the network. The technical phases of the process are described, the main files generated by Bitcoin Core are analyzed, and the effects of the parameters txindex, prune, dbcache, maxmempool, and maxconnections are empirically studied. System resources during the block download (IBD)…
The full abstract on arXiv