System Requirements

Note

The goal of this page is to set expectations and get you to understand what is required to run dfuse, at different scale.

The dfuse platform is extremely elastic, and supports handling networks of varied sizes and shapes. It is also heavy on data, so make sure you have a good understanding of the different data stores, artifacts and databases required for dfuse to run.

The deployment efforts will be proportional to the size of history, and the density of the chain at hand.

Network shapes

This document outlines requirements for different shapes of networks

Local development chain

dfuse for EOSIO can be booted on your local computer with minimal overhead on top of the blockchain node (which itself can be trimmed).

It will boot all the components and APIs in a single process.

You can control the amount of RAM and storage used by the nodeos process with these config.ini flags (in both producer and mindreader folders):

chain-state-db-size-mb = 50
chain-state-db-guard-size-mb = 5
reversible-blocks-db-size-mb = 4
reversible-blocks-db-guard-size-mb = 2

These will keep your nodeos node’s RAM consumption under 200MB. Twice that if you run both a producer and a mindreader node.

A fresh start dfuseeos on a blank chain (after a fresh dfuseeos init and/or dfuseeos purge) will consume around 700MB. However, plan 4GB of RAM for growth, increase in the number of transactions in blocks, and for use during an extended period of time.

Persistent chains

In order to scale easily, you will want to decouple components that run in a single process in local instance deployments.

The storage requirements will vary depending on these metrics:

  • The length of history: which will affect the quantity of search indexes that need to existe in dstore storage, and being served by search-archive nodes, thus the amount of RAM
  • Throughput in actions: Actions in EOSIO are the smallest units of execution to produce meaningful data, transaction overhead becomes negligible once you have 2-3 actions in a transaction. A single token transfer generally has 3 actions.

The CPU/RAM requirements will depend on these factors:

  • High Availability: highly available deployments will require 2 times the resources listed in the following examples, as a general rule.
  • Throughput of queries: the dfuse platform is built for horizontal scalability, the more requests per second you want to fulfill, the larger the deployment, the more CPU/RAM you will need to allocate to your cluster.

Note

These stats are from July 2020, but chains of similar density and similar age will behave similarly.

Medium scale chains

CryptoKylin:

  • https://www.cryptokylin.io/
  • Age of the chain: 114M blocks (2.00 years old)
  • TPS: ~5 tps
  • This is a long running chain, with a low volume of transactions.
  • Search components:
    • Indexes size: 230 GB
    • Object Storage for those indexes, compressed at approximately 18%.
    • archive node for full history:
    • CPU/RAM: 4 vCPUs and 12GB RAM. Double that for HA.
    • Live SSDs to host 230 GB uncompressed indexes
    • live nodes:
  • trxdb size: 560 GB
  • fluxdb size: 27 GB

Large scale chains

EOS Mainnet:

  • https://eosq.app
  • https://bloks.io
  • Age of the chain: 131M blocks (2.08 years old)
  • TPS: on June 30th 2020, an average of 819 action traces per second. Throughout the history of the chain, there was not always that volume of data.
  • Search components:

    • This includes FILTERED search data. This excludes eidos mining. See filtering query below.
    • Indexes size: 5.2TB
    • Stored as compressed zstd indexes in Object Storage (dstore), 19% of the total size.
    • CPU/RAM for archive nodes: around 10 vCPUs, 100GB RAM (double that for HA)
    • An uncompressed copy of those 5.2TB indexes on live SSDs.
    • This assumes around 10 new incoming requests per second.
    • NOTE: In search, many users use long-lived streaming connections. This could still mean thousands of in-flight requests.
    • CPU/RAM for live nodes: around 4 vCPUs, 24GB RAM (double for HA, triple, quadruple to scale real-time reads)
    • In-flight requests will affect the live deployment sizing.
    • CPU/RAM for indexer nodes: around 2 vCPUs, 8GB RAM.
    • If you have multiple search tiers with different blocks sizes, it could mean 2 of these.
  • Size of trxdb (this data includes all data, unfiltered, including “spammy” transactions): 15.1 TB

  • Size of fluxdb key/value store (unfiltered): 454 GB

    • This does include all state changes for the whole history of the chain, rounded at the block height.

The estimates for search indexes imply this search filtering:

dfuseeos start
      ...
      --search-common-action-filter-on-expr=""
      --search-common-action-filter-out-expr=account == 'eidosonecoin' || receiver == 'eidosonecoin' || (account == 'eosio.token' && (data.to == 'eidosonecoin' || data.from == 'eidosonecoin')) || account == 'eosiopowcoin' || receiver == 'eosiopowcoin' || (account == 'eosio.token' && (data.to == 'eosiopowcoin' || data.from == 'eosiopowcoin'))