Storage Provider Survey

2021-08-31 Asia SPWG (large miners)

2 message : 1 message → 5 : 2

  • IPFSUnion : 2 message
    • hardware resource is the main factor here. from a technical development perspective, the solution that doing less computation and uses less GPU is always better for the longer term
  • Lengbo StorSwift: 2 message
    • want to know the precise gas usage between two, if overall gas usage is the same, then 2 messages is preferred
  • 1475: 1 message
    • will only consider starting to upgrade cc sector when has limited hardware resources(disk storage..). When that happens, it also means they won't seal as many sectors as they do now which will free up more GPU resources - then the heavier GPU usage for 1 message protocol is fine
  • ipfsmain: 1 message
    • if both introduce new crypto proof
  • ipfsforce: 2 message
  • Charles: 2 message
    • Is focusing on making deals so prefer to save GPU for deals
      • fatman13: 2 message<

Other Consideration

  • optimization potential: 1 message protocol would be easier to optimize as it doesn't have the enforced randomness wait
  • from an ecosystem PR perspective, less gpu consumption → greener relatively (environmental perspective)
  • Access to GPU - hard to get GPU in Canada
  • Will you use the new upgrade CC sectors?
    • Mostly mentioned not right away
    • constraint by clients ask, here clients = the ones who buy the power and will likely need to seal new sectors for them for profit sharing purpose
    • for storage clients, the time saving from the new upgrade cc comparing to seal in the new sector is not appealing, almost can be neglected
  • 1 asked if can support both options and got some push backs, as others think core devs should be decisive based on benchmark data and provide one clear path to the Community

Open Questions

Note:

  • This group is more technical so they all would love to know more details of both proposals, including new proof being introduced what's the difference, and the actor implementation change.
  • almost all of them do not think 2 message approach is more error prune

2021-09-01

2 message : 1 message → 11 : 3

Open Questions:

  • how much is the collateral for the 2 message protocol
  • For both protocol, deal activation seems to be v fast - how is network's security generated?
  • one of the pro of 1 message protocol is the deal activation time is shorter, however, Is deal activation really the sooner the better?
    • if a SP can provide 1.5hr deal activation service for X days till CC sectors are all upgraded, that means then they would have to provide worse and slower service with longer deal activation since they would have to onboard deals in new sectors with a longer sealing process, from a service perspective this is a retrogress.
      • that being said, SP may have to always keep sealing CC sectors so they can provide consistent service. This is bad as SP should've be able to only need to perform sealing and onboard deals in new sectors instead of have to do sealing + upgrading.
  • CE: what should be the new IP?
    • if it's going to be max(old IP, new IP) it's going to be irrational for SP to upgrade older CC sectors as the IP was way higher back then, and assuming most of deals will be verified deals and require 10x collateral → sealing new sectors with new IP would lock up less collateral. SP would just let the garbage CC sectors to expire and get collateral back in that case.
  • Gas cost: we are saying the gas cost will be very comparable for two protocols, does this statement stay true if base fee spikes?
  • Known out-of-scope for this fip, but if onboarding deals can be this fast, can we do this with deals sectors too in the future? For use up un-filled sector space maybe?

Actor/Lotus implementation considerations

  • lotus: if go for 1 message protocol:
    • should be able to do the computation on a worker rather than have to be on the main miner
    • should have wdpost worker - otherwise its un-usable for SP that runs everything on one machine (should apply to 2 message protocol as well)
  • lotus+ actor Expiration:
    • for both protocol:
      • what happens if one of the deals is expired when the message is landed on chain
    • for 2 message protocol:
      • what happens if the sector is expired between message 1 and 2
      • what happens if one of the deal is expired between message 1 and 2
      • what happens to the collateral
  • actor + lotus: allow collateral (focus on 2 message collateral in this case) to be returned to an worker/control address to avoid gas cost of lotus-miner actor withdraw
  • lotus: be able to do snap deal by default when there exists CC sectors
  • lotus: local sector status update