LogoLogo
v2.x (Stable)
v2.x (Stable)
  • What is Boost?
  • Features
  • Components
    • boostd
      • Repository
      • GraphQL API
      • JSON-RPC API
      • SQLite metadata database
    • Local Index Directory
      • Index types
      • Dependencies
      • Initialisation
    • boostd-data
    • YugabyteDB
    • booster-http
    • booster-bitswap
    • libp2p Protocols
  • Hardware requirements
    • YugabyteDB
  • Installation
  • New Boost Setup
  • Configuration
    • UI Settings
    • HTTP Transfer limit
    • Deal Filters
    • Remote CommP
    • Legacy Deal configuration
    • HTTP indexer announcement
    • Manual Publish Storage Deal Message
  • Monitoring
    • Setting up a monitoring stack for Boost
  • Storing data on Filecoin
  • Retrieving data from Filecoin
    • HTTP retrieval
    • Advanced Configuration of booster-http
    • Bitswap retrieval
  • Backup and Restore
  • Tutorials
    • How to upgrade from Boost v1 to Boost v2
    • How to re-index unsealed pieces that are flagged by LID in Boost v2
    • How to upgrade from v2.0.0 to v2.1.0
    • How to upgrade from v2.1.x to v2.2.0
    • Start and stop Boost processes
    • How to store files with Boost on Filecoin
    • Using filters for storage and retrieval deals
    • Migrate from Lotus to Boost
    • How to onboard data using DDO deals
  • Troubleshooting
  • Experimental Features
    • FVM Contract Deals
    • Direct Deals
    • Data Segment Indexing
  • FAQ
  • Need help?
Powered by GitBook
On this page
  • General hardware requirements
  • CPU
  • RAM
  • Disk
Edit on GitHub
Export as PDF

Hardware requirements

Hardware requirements for Storage Providers running Boost

The hardware requirements for Boost are tied to the Lotus stack in a Filecoin SP deployment.

Depending on how much data you need to onboard, and how many deals you need to make with clients, hardware requirements in terms of CPU and Disk will vary.

General hardware requirements

CPU

A miner will need an 8+ core CPU.

We strongly recommend a CPU model with support for Intel SHA Extensions: AMD since Zen microarchitecture, or Intel since Ice Lake. Lack of SHA Extensions results in a very significant slow down.

The most significant computation that Boost has to do is the Piece CID calculation (also known as Piece Commitment or CommP). When Boost receives data from a client, it calculates the Merkle root out of the hashes of the Piece (padded .car file). The resulting root of the clean binary Merkle tree is the Piece CID.

RAM

128 GiB of RAM is needed at the very least.

Disk

Boost stores all data received from clients before Piece CID is calculated and compared against deal parameters received from clients. Next, deals are published on-chain, and Boost waits for a number of epoch confirmations before proceeding to pass data to the Lotus sealing subsystem. This means that depending on the throughput of your operation, you must have disk space for at least a few staged sectors.

For small deployments 100 GiB of disk are needed at the very least if we assume that Boost is to keep three 32 GiB sectors before passing them to the sealing subsystem.

We recommend using NVME disk for Boost. As Dagstore grows in size, the overall performance might slow down due to slow disk.

Previouslibp2p ProtocolsNextYugabyteDB

Last updated 1 year ago