Resilience of the Filecoin Network

In this article, we discuss the resilience of the Filecoin network.

TLDR

• Filecoin is resilient to significant power drops because of anti-fragile mechanisms built into the protocol design.
• The mechanisms provide strong economic incentives for new storage providers (SPs) to join and existing SPs to stay in the network after power loss.
• Specifically, after significant power loss, power onboarded onto the network will gain outsized rewards due to a sharp increase in reward concentration.

Introduction

Filecoin is a decentralized data storage network that stores humanity’s most important information. SPs provide storage capacity to the network for a defined contract period and receive block rewards in compensation. To ensure the contract is honored, collateral in FIL tokens is required. The collateral amounts and block rewards received change based on the type of power being onboarded (empty capacity vs. real data).

What does resilience mean? Merriam-Webster defines resilience as the ability to recover from or adjust easily to misfortune or change. Resilience is relevant to many aspects of a cryptoeconomy like Filecoin; examples include 51% attacks, Sybil attacks, and Finney attacks, to name a few. In this article, we focus on a specific aspect of resilience: the Filecoin network’s resilience to power shocks. In Filecoin, power refers to the amount of storage and data the network serves. Power shocks can result from SPs dropping out of the network for various reasons.

Let us begin by precisely framing questions related to resilience that we would like to answer quantitatively: How will the Filecoin Network respond to the power-loss scenario where a significant portion of SPs either:

1. Leave gracefully, expiring power off the network after completing their storage contract
2. Terminate their storage contracts suddenly before the end date, causing a shock and a sudden loss of power to the network.

Modeling

How can we model the network’s resilience in a principled way and ensure that our forecasts follow the cryptoeconomics mechanisms of Filecoin? Our approach is to use our Agent-Based Model (ABM) of the Filecoin economy. ABMs consist of an environment and agents which interact with the environment. In this ABM, the environment implements the cryptoeconomic mechanisms of Filecoin (such as locking, vesting, minting, and the interaction between these aspects of token supply) and agents model SPs. Refer to this article for more details about our Filecoin ABM.

To understand resilience, we build agents that model three types of SPs in the Filecoin network:

1. SPs that onboard only committed capacity (CC) sectors
2. SPs that onboard only deal sectors (Fil+)
3. Mixed SPs which onboard both CC and deal sectors

Each SP can behave in one of the following ways:

1. Dollar Cost Averaging Behavior — here, the agent will onboard and renew a constant, preconfigured amount of power until the termination date, after which the agent will leave the network (either through power expiration or termination).
2. Adaptive ROI behavior — in this mode, the agent will use forecasts of FIL-on-FIL Returns (FoFR) to determine how much power to onboard if the forecasted FoFR is greater than a configured threshold.

With these agents and the environment specification, we configure several counterfactual scenarios representing different network states from which a power shock scenario can occur. Define network state as a distribution of the type of SPs that operate in the network. Each network state can lead to varying outcomes since each has unique raw-byte and quality-adjusted power profiles, leading to different minting, pledge, and FoFR dynamics. The starting network states we test are:

1. An even distribution of CC, FIL+, and Mixed SPs
2. An even distribution of CC and FIL+ miners only
3. A 70/30 split between FIL+ and CC miners, respectively (FIL+ Skewed)
4. A 30/70 split between FIL+ and CC miners, respectively (CC Skewed)

From these points, we simulate the counterfactual cases where the network power drops either 30% or 70% from its current value. Two variants of network power drops are considered, a) the power is expired gradually from the network, or b) the power is terminated from the network.

To recap our simulation matrix, we simulate a 30% or 70% network power drop for different network starting points through a gradual sector expiration process or an immediate termination event. We simulate each configuration and record network key performance indicators (KPIs).

Results

Fig. 1 shows various network KPIs when power gradually expires, and Fig. 2 shows the network KPIs when power terminates suddenly.

Some nuances differentiate the gradually leaving and the sudden termination cases, but in both power loss scenarios, we observe that the network power begins to recover after the termination event. Recovery occurs because FoFR increases due to an increase in reward concentration.

It works like this: After termination, network power declines rapidly but has the inverse effect on normalized rewards per QA sector, which increases. Combined with a declining pledge, a concentration of rewards occurs, resulting in a sharp increase in FoFR. This combination of network conditions gives high FIL-on-FIL returns for participants who stay with the network. Our simulation reflects this phenomenon — here, the agents (rational actors) observe the high FIL-on-FIL returns and take advantage of the situation to onboard more power. This enables the network power to recover.

More details, additional scenarios, and discussion can be found in this detailed report.

Filecoin Network KPI’s under several conditions of SP’s leaving the network. The dashed vertical line denotes the start of the simulation, and the dotted vertical line denotes the date at which power starts leaving the network. A baseline case using a DCAAgent to simulate constant onboarding is also simulated to provide a basis of comparison (black dotted line).

Filecoin Network KPI’s under several conditions of SP’s terminating from the network. The dashed vertical line denotes the start of the simulation, and the dotted vertical line denotes the date at which power starts leaving the network. A baseline case using a DCAAgent to simulate constant onboarding is also simulated to provide a basis of comparison (black dotted line).

Conclusion

In this work, we have explored the resilience of the Filecoin network to network power shocks. We conclude that the network is resilient because of rewards concentration, which boosts the income of remaining miners, and anti-fragile because of the baseline recovery mechanism.

If you want to read more about CryptoEconLab and our work, visit our website at cryptoeconlab.io or follow us on Twitter.

This article was made for informational purposes only. CryptoEconLab does not provide legal, tax, financial or investment advice. No party should act in reliance upon, or with the expectation of, any such advice.