Up until now, we have primarily done real world metrics gathering and haven’t had a traditional load testing suite for retrieval. Over December 2022 and January 2023, we will be building this suite.

Repository

https://github.com/filecoin-project/retrieval-load-testing

Protocols

• Start with testing http piece retrieval since it should be the lowest lift to test (many existing tools). By showing the reliability of http piece retrieval, SP’s will have a clear retrieval method to show to prospective clients.
• Follow http testing with bitswap. We have made a multi-month effort to roll this out and gives us an end to end story with IPFS. Need to show it is reliable software to get adoption.
• Lastly, and optionally, move to graphsync. More complex to set up since the dag of the data matters.

Setup

• Test a real, semi-isolated miner on mainnet. We will test on Boost SP 0 to start. We should think of this as — what could we as Boost SP 0 pitch to an enterprise customer.
• Don’t worry about isolated environments that mock out the sealer – we want to know about problems in the sealer. We also want to understand tertiary effects of load – if a real retrieval forces a provider to miss WindowPOST, we’d like to learn about that.

Preparation

• We need to get data onto Boost SP 0. We could do this a couple ways — go get some FIL+ and make some deals, or just put a retrieval filter to only accept some wallets, and make non-FIL+ deals.

  • Script on boost-sp-0 to:
    • Generate 10 random files of 20GB (32GB piece),
    • convert to CAR and then make offline deals for them
    • import the data
    • now available in load testing repository
  • Would be nice to get a real data set like Wikipedia on there (nice to have)

• Working Deals: 1.

    | UUUID | a3d6c841-b9f9-4058-94d6-1e0bdca1b275 |
    | --- | --- |
    | Data CID | bafykbzacebmouqhg33afbz2sb57ivdfxpw3si6w7q7or32k53axuhhvi3myms |
    | PieceCID | baga6ea4seaqimqd4qmejl3xlvcuwrtkl4drvxwv4dpvis4hiwt5wa5e2645eyey |
  

    | UUUID | 7ae33dca-6847-4252-8ad0-2b9efadf88be |
    | --- | --- |
    | Data CID | bafykbzacedyjregukngy4nf6uyxzcaaxhwdkgi2ybq3xheusygkgpvifkpqbu |
    | PieceCID | baga6ea4seaqllfctnxlckjeeakeyca2h5vxtzthycawcnkah5igxw7i6ygcn2oy |
  

    | UUUID | 710a018e-70fa-43e3-91f5-9c07e1fa8e88 |
    | --- | --- |
    | Data CID | bafykbzacecdfbsdxbmfiu272sa364fzphjydzws7xlhyklihn2t6dpvgnpdrc |
    | PieceCID | baga6ea4seaqa2kbljopadmcwp2swghamozhr5yypekryoeaysnje5ue7wj3d6oi |
  

    | UUUID | 639817bc-dee1-4764-b624-ae99735db729 |
    | --- | --- |
    | Data CID | bafykbzaceb4gqljh5wrijjincngznnrw4f6hwjfpei4evvcwhhgh4jegjb4sy |
    | PieceCID | baga6ea4seaqjv5cz75va35qaiju4h5a4guyqbahnyubqwrocqulkkek56cguwcq |
  

• Collateral for Boost SP 0

  • Resolved: Sufficient for now

• Booster HTTP public setup

  • Setup Nginx
  • Serves /piece for proxy to booster-http
  • Service /rawpieces to folder with static files as pieces
  • SSL with self-signed cert and basic auth, cert and creds in 1Password, ask @Hannah Howard

• Running tests directly from boost-sp-0 with traffic shaping, 1Gbps and 10Gbps. Example TC commands (1Gbit/s):

sudo tc qdisc add dev lo root handle 1: htb
sudo tc class add dev lo parent 1: classid 1:20 htb rate 1000mibit
sudo tc qdisc add dev lo parent 1:20 handle 20: sfq perturb 10
sudo tc filter add dev lo parent 1: protocol ip prio 1 basic match 'cmp(u16 at 0 layer transport eq 443)' flowid 1:20

• Would be great to get client wallet w/ funds (nice to have)

Workloads

• Generally want to get more input from actual clients on this. As a starting point–