How does FVM comprehensively unlock innovative use cases for Filecoin?

Source: Siddarth, IOSG Ventures

What is FVM and why do we need it?

Protocol Labs has outlined their three-phase plan for a decentralized internet:

1. Establish the world's largest decentralized storage network.
2. Introduce and protect human data.
3. Bring retrieval and computation capabilities to the data to build scalable applications.

Filecoin has already achieved the first step, becoming the largest decentralized data custodian with over 661.54 PiB of data storage. The Filecoin storage market continued to grow in Q4 2022, with active transactions increasing by 117%, a year-on-year growth of 1,798%.

The next step for Filecoin is to help introduce and protect human data, which will be an ongoing process. To achieve the third step, new infrastructure needs to be established to assist in data retrieval and computation; otherwise, Filecoin will merely become a pile of hard drives archiving data around the globe.

This led to the birth of the Filecoin Virtual Machine (FVM). By integrating the virtual machine into the Filecoin protocol, developers can create decentralized applications (dApps) based on the Filecoin storage network, execute smart contracts in a secure and reliable manner, and create additional value layers on existing Filecoin data.

FVM is designed to execute smart contracts on the Filecoin network. When a smart contract is deployed, it is compiled into WASM bytecode. This allows developers to create useful applications that behave according to immutable rules, unlocking various use cases such as bringing all of Filecoin's markets on-chain, permanent storage, DataDAOs, and more. While all of this is promising, we have seen many chains fail without a good developer experience and a lack of developer markets.

Filecoin decided to launch and support FEVM. When an EVM is deployed to FEVM, it is compiled into WASM, creating an actor instance in FEVM to run the EVM bytecode. Then, user-defined FEVM actors can interact with the Filecoin network through built-in Market and Miner APIs.

This is a critical step for Filecoin because EVM-based development has proven to provide a good developer experience and has a large existing developer community. By providing a secure and reliable environment for executing smart contracts, Filecoin VM helps unlock the full potential of the Filecoin protocol, bringing new innovative application scenarios for decentralized storage and computation.

Use Cases Unlocked by FVM

While many different use cases can be built using FVM, Protocol Labs and the FVM team have proposed a "Request for Startups" list that they believe is crucial for the prosperity of the Filecoin ecosystem. Here are some highlights from that list:

• DataDAOs?
• ML model storage and enhancement [via DataDAOs]??
• Storage gateways [via DataDAOs]??
• Pay-per-use [via DataDAOs]
• Gaming [via DataDAOs]???
• Social [via DataDAOs]???
• Decentralized science [via DataDAOs]???
• Tracking and reducing Filecoin's carbon footprint?
• Hardware collateral lending
• Permanent storage??
• Storage automation (replication and repair)??
• Trustless FIL + notary??
• KYC and proof of claims
• Decentralized data aggregators??
• Access control

According to Protocol Labs, the asterisk indicates the importance of the project on the Filecoin network. Readers may have noticed the repeated mention of DataDAOs; what are they and why are they important?

DataDAOs

The saying "data is the new oil" has been circulating in recent years. The largest internet companies in the world have always considered data their most valuable resource, whether for internal use (market/user insights) or external data sales. Why is data valuable? Data is only valuable when it can generate insights, and generating insights requires computation and analysis of the data itself.

In the current state of Filecoin, monetizing data on-chain is impossible because the storage protocol is peer-to-peer, transactions are conducted off-chain, and settlements occur on-chain. Monetizing data requires infrastructure to establish access control systems, subscription payment systems, data enhancement, packaging, etc.

This is what FVM can unlock, adopting a decentralized, community-centric model. DataDAO is a type of DAO whose mission revolves around the protection, generation, enhancement, and promotion of datasets deemed valuable by stakeholders.

Each stakeholder in a DataDAO can now be incentivized in various ways: SPs are responsible for storage and preservation, replication workers can ensure local and fast availability, data providers can now receive compensation for the sale of their data, data experts can participate in data packaging, and ML engineers can be compensated for providing models that run on encrypted data, thereby increasing the value of the data.

We can view data as a value spectrum:

1. Raw data (least valuable)
2. Packaged data
3. Data computation
4. Verifiable insights (most valuable)

In the state prior to FVM, raw data and packaged data could be uploaded to Filecoin but could not have access control on-chain. Now, data computation (e.g., AI/ML models, NFT creation, etc.) can be performed using FVM. However, in the current state of Filecoin, complex and heavy computations cannot run on-chain because Filecoin nodes currently lack sufficient computational power. This indicates a clear scope for off-chain Dcompute platforms that interoperate with Filecoin.

Solutions like Bacalhau are dedicated to achieving data-based computation and have close interoperability with Filecoin. Bacalhau is a relatively new project that can make the Filecoin network more valuable by allowing data stored in Filecoin itself to become more valuable.

DataDAOs do not need to cover the entire data value chain in their initial state. While this may be the ultimate goal of DataDAOs, many DataDAOs have already begun working in specific verticals or use cases. Some notable examples include:

● Lagrange DAO: A DAO for data value realization and decentralized science (DeSci). It provides a space for data sharing and analysis for DeSci.

● GlacierDAO: A DAO that replicates an open Git repository containing public interest code. Additionally, it allows users to pool funds to finance the replication of these repositories on the Filecoin network.

● SPN DAO: A DataDAO that enables consumers to convert credit card transaction data into assets, allowing them to directly control the use and monetization of their data.

While DataDAOs currently focus on data collection and management, computation of data and DCompute infrastructure will be needed to enhance the value chain. Projects that are already working on this and focusing on Filecoin include:

● Bacalhau: Bacalhau is a platform that achieves fast, efficient, and secure computation by running jobs where data is generated and stored. By running arbitrary Docker containers and WebAssembly images as tasks, users can streamline existing workflows without extensive rewrites.

● Shale: Shale is working to bring cloud computing to Filecoin, enabling storage providers to leverage existing storage capacity for computation and directly compete with other cloud storage providers like AWS and Google Cloud. Public users will have the opportunity to rent computing instances from storage providers and access Filecoin+ storage transactions through local networks. This is a solution similar to Akash Network and StackOS, but it places more emphasis on computing with Filecoin data.

DataDAOs are a unique use case that can only be unlocked in a decentralized manner end-to-end through the Filecoin network.

Addressing Existing Issues

Filecoin was initially just a storage protocol, and an evolving protocol inevitably has many issues that need to be addressed, such as:

1. How do I fix or retrieve my data if my SP is penalized for not storing data?
2. I have high-quality content that needs to be delivered to my distributed website using a caching layer, but all my transactions with SPs are conducted offline.
3. If I have enterprise-level data, how do I implement access control on my data on Filecoin? There are many other questions.

Filecoin is working to address these issues, such as FVM and Retrieval Markets. FVM unlocks the possibility of creating and incentivizing data replication on Filecoin, so even if one SP fails, data can be retrieved or repaired from other nodes.

Retrieval Markets will help create a decentralized CDN network, which is very beneficial for Web3 social and gaming projects. Alternatively, an access control platform with end-to-end data encryption can be established using FVM, with projects like Medusa making efforts in this direction while also considering additional interoperability layers with other chains.

Strengthening the Demand for DCompute Infrastructure

In distributed data storage systems like Filecoin, performing computation can be somewhat challenging because data may reside on many different nodes far apart. Aggregating data and then computing it would completely undermine the purpose of a decentralized storage system.

As the world moves towards training AI models on large datasets, storing large datasets on Filecoin has enormous economic value (due to low costs), but currently, building training models on these systems is very difficult.

ChatGPT (GPT-3) was trained on a machine with 1000 V100 GPUs (each GPU has a computational speed of about 147 TFLOPs), with an average cost of about $10 million.

Stable Diffusion was trained on 256 A100 GPUs, costing nearly $600,000.

This requires immense computational power, while the current minimum requirements for Filecoin nodes are 8-core GPUs and 128 GB RAM. Filecoin alone cannot cope with the computationally intensive internet era we live in.

There are two ways to address this issue:

1. By setting the minimum computational requirements for storage very high, making Filecoin a computational powerhouse. This is not a transformation that can happen overnight. It must be a gradual process.

2. Outsourcing the computation of data on Filecoin to third-party Dcompute platforms, such as Bacalhau and Shale (Filecoin-centric), or other solutions like Akash Network, StackOS, etc.

Another issue is that many training models have already been built, and if data needs to be migrated to Filecoin, the computation network must also support existing models built using TensorFlow, Ray, etc.

In the short term, I believe decentralized cloud computing networks like Shale and Akash will prevail in the market due to ease of deployment and lower overhead for developers, while in the long term, Filecoin must also strive to become a computational powerhouse by upgrading existing resources or finding more efficient decentralized computing methods.

Rex St.John proposed a longer-term solution in his article "Harvested Compute and Pervasive AI: The Protocol Labs Bull Thesis," suggesting that the Filecoin network should focus on supporting computation using Raspberry Pi and Nvidia Jetson, as the number of these units is already available in the market and can more consistently support node bootstrapping.