- Abstract
- Activities of Storage Providers
- Fat SP Thesis (status quo)
- Drawbacks
- Thin SP Thesis
- New SP Economy
- How do we get there?
- Drawbacks
Abstract
When Filecoin launched, the reference implementation for Storage Providers (SPs) implemented all of the relevant activities that a Storage Provider should do in order to provider their storage services to the Filecoin clients.
After ~20EiB of capacity on-boarded to the network and over 2 years since launch, this blogpost summarizes the main activities of SPs and evaluates two thesis for SPs: a Fat SP thesis, where SPs are assumed to perform all of their activities and a Thin SP thesis, where SPs outsource most of their activities to service providers and focus on storage.
Activities of Storage Providers
There are several key activities that a Storage Provider does today:
Storing data | Storing data on disk durably and over time. It is the main service for which SPs are paid for when making a deal. |
Deal making | Looking for clients to sell storage deals. |
Pledging collateral | Depositing FIL as a guarantee of correct execution of storage services. Generally speaking, collateral is proportional to the storage and duration they on-board |
Sealing | Preparing storage for participating in Filecoin mining by sealing sectors (either CC or with deals). It is a one time expensive operation that requires expensive computation. |
Activity | Description |
Proving storage | Generating proofs to the network showing that data is being stored and space is dedicated to the Filecoin network. It is repeated once a day and it is divided in two steps: looking for data (essentially looking for some requested random bits across all storage) and generate a proof (expensive operation that requires GPUs). |
Serving data | Serving data when requested by users. |
Producing blocks | Choosing transactions in the mempool, creating blocks and propagating them. |
Fat SP Thesis (status quo)
For most Storage Providers, core devs and protocol designers in our ecosystem the Fat SP thesis has long been the status quo.
- SPs perform most of activities in-house.
- A single client or a modular SP stack (e.g. lotus) that can perform all of the activities
If the community continues to focus on the Fat SP thesis, then the main goal of core devs and protocol designers must be to optimize the software and the protocol to reduce the cost of running such a complex operation.
Drawbacks
Each of these activities require very specialized operations which only very few SPs can afford to do well at a low cost. To add more context:
- “Sealing” and “Proving Storage” require proper compute infrastructure such as special GPU set-ups.
- “Storing data” requires proper storage infrastructure such as network storage, HDDs, fault recovery mechanism.
- “Pledging collateral” requires proper financial risk management and tooling for taking loans, reducing volatility of future rewards. In other networks such as Ethereum, most validators get liquidity from external liquidity providers (e.g. Lido, DiVa, RocketPool).
- “Serving data” requires proper network infrastructure to have proper CDN-quality setups
- “Mining blocks”, and in particular choosing transactions, has revealed to be a very complex operation that in other networks (such as Ethereum’s) validators outsource the making of blocks to external block producers.
Finally
- Steep requirements: Only few storage providers can be “fat” providers, the cost of such setup can be pretty steep. SPs to grow expertise in several areas of their business:
- Running proper infrastructure for: compute (GPUs, large scale compute), storage (HD, network storage), network (CDN-like infra)
- Running proper business units: marketing (for finding clients), etc
- High capital cost: The Fat thesis requires a high capital cost on the SPs.
- Complex protocol/setup: Optimizing for Fat SPs makes our stack complex (e.g. lotus miner needs to support all features of a Fat SP)
Thin SP Thesis
As of 2022, new entities appeared in the ecosystem providing services to SPs by effectively lowering some costs of their operations. There are new SPs appearing in our ecosystem that are already starting to delegate core operations such as sealing or pledging collateral.
- Storage Providers perform only storage activities in-house and delegate most of their activities to an economy of service providers. The SPs should merely focus on storing data.
- There is not a single client that can do all, but there are different clients for different parts of the operations.
- The Thin SP thesis turns the capital costs of SPs into operational costs.
There are several priorities that would shift if the Thin SP Thesis is a thesis that the community starts to adopt more, starting from making new specialized clients for the different activities and focusing on making these activities easier to delegate. Such as making sealing easy to delegate
New SP Economy
Activity | Service | Description of service | Players | Status |
Sealing | Sealers
(or Sealing-as-a-Service) | Sealers seal sectors (with or without deals) and sell them to SPs. Sealers can also generate proofs. | FilMine, Aligned.co | Supported with working services |
Deal Making | Data Preparers (DP) | DP generate single large deals that can fill an entire sector, and auction their deals. This massively reduces the cost of running business operations for looking for deals. | Estuary, Web3.Storage, BigD.Exchange | Early stage |
Proving storage | see ^ | see ^
Highly likely not smart to delegate. | see ^ | Supported with working services only for Sealing-as-a-Service
Not supported for WinningPoSt or WindowPoSt |
Storing data | SP | This is the only activity not outsourced | n/a | |
Pledging collateral | Liquidity Providers (LP) | LPs provide liquidity to SPs in various ways: liquid staking derivatives, lending pools, etc. | PalladiumX
Anchorage
CoinList
FilMine,
Glif | |
Serving data | Retrieval Providers (RP) | RPs provide access to data to final users by keeping an extra copy and relying on retrieval from SPs in case of faults. | Filecoin Saturn, Estuary, Retriev | Although Retrieval Networks are happening, there is no plug in solution for SPs |
Mining blocks | Block Producers | BPs make blocks of transactions. (Similar to Ethereum) | ? | Not developed |
How do we get there?
Most services will organically show up without our effort since they will end up being the most efficient solution (e.g. Sealing-as-a-Service or Lending apps).
However, the core implementations and the protocol could ease this transition by making it easier for these new players to appear.
- For example, the lotus team had a large effort focusing on better standard and interfaces to make Sealing-as-a-Service easier to plug in.
- Another example, will make it easier for Sealing-as-a-Service to optimize their sealing pipeline to reduce their costs.Inactive CC for a CC market
- Another example, making allows to verify if a subpiece is included in a deal, which can enable a market for data preparers, where clients do not need to trust DPs.Verifiable Off-chain Data Aggregation
- Finally, the and FVM will make it easier to have lending markets or data preparers services.Filecoin storage/data programmability
Drawbacks
- In the Fat SP thesis, the SP do not depend on anyone for participating in the network, however in the SP thesis, service providers could stop offering their services to specific SPs.
- If there are no multiple opportunities (and competition), then a single group could have too much power in the network.