TL;DR
-
As Gen-1 Node Providers (NP) are nearing the end of the 48 month period, the currently accepted proposal for an updated topology and remuneration (proposal #127044) was re-evaluated by NPs. We are now presenting an updated proposal.
-
The goal of this updated proposal is to:
- Decrease inflation due to NP rewards (33% reduction in Gen-1 rewards, overall 3.2% reduction in inflation),
- Keep as many Gen-1 nodes in top tier data centers online as possible (to provide capacity for IC’s growth)
- Simplify the remuneration scheme.
-
We would like to propose an approach that not only balances Gen-1 NP rewards and network inflation, but also improves network decentralization without reducing the total node count. We ask that the community vote to accept this proposal. Voting to accept would mean:
- The rewards for Gen-1 nodes operating beyond the initial 4-year period would be reduced by 33% from current levels, recognizing that while the initial purchase/setup costs have been paid off, operational costs have risen.
- This revised reward scheme would be in place for a minimum of 2 years, providing Gen-1 NPs with the stability needed to make necessary commitments to their service providers, such as data centers and ISPs.
- NPs will be limited to operating a maximum of 42 nodes, in line with the target topology.
- Any excess nodes from the 42-node cap can remain on the IC. While reassignment to a “New” NP is mandatory, these nodes may continue in the same or a different data center and/or country. A “New” NP can also be an existing NP with fewer than 42 nodes. No NP will receive rewards for more than 42 nodes, and the new NP will receive the same 33% reduced remuneration for the acquired nodes. New NPs not yet listed in the NNS will be accepted through the self-declaration process and respective proposals.
- NPs taking on these excess nodes will be allowed to onboard them without rejection due to target topology restrictions, as this is not an addition of new nodes but a reassignment that maintains the total node count and enhances network decentralization and utilization.
Background
On November 12, 2023, a motion proposal (#125549) was published and accepted where the community agreed on a target topology. The target topology defines the number and sizes of subnets based on anticipated demand and sets decentralization targets for each subnet using a Nakamoto coefficient or subnet limit.
Following this, on December 21, 2023, a forum post was made to update Gen-1 node provider remuneration in line with the new target topology and the fact that these machines were approaching the end of their initial 4-year period. The 4-year timeframe is significant because, under the original model, it was estimated as the time needed for the initial purchase/setup costs of these nodes to be fully amortized. On January 19, 2024, a motion proposal (#127044) for this updated remuneration scheme went live for voting and was accepted.
Although the community, including NPs, adopted proposal #127044, several challenges and concerns were raised after running simulations and discussing the results during the IC Lab for NPs. The primary concerns were that these changes could make node operation economically unviable for nodes in top-tier data centers and may require the removal of servers from the network.
These concerns were shared with DFINITY during the last day of the Node Provider ICP Lab (May 16, 2024), and following that, a group of NPs worked on an updated proposal. After incorporating feedback, we are now ready to share it with the community for broader input.
The rest of this forum post outlines the proposed updates, explains the need for revising motion proposal #127044, addresses inflation caused by Node Provider rewards, and concludes with a summary of the concrete outcomes if this proposal is accepted.
Proposed Updates
We would like to propose an approach that not only balances Gen-1 NP rewards and network inflation, but also improves network decentralization without reducing the total node count.
Node Provider Remuneration
Since NPs’ initial setup costs are typically recovered over 48 months, the monthly rewards are reduced accordingly, though not as drastically as previously suggested. Proposal #127044 reduces rewards by around 75%, which makes it extremely difficult for node operators to continue running nodes.
To clarify, the rewards structure is designed so that within approximately 4 years, the upfront investment (CAPEX) is fully paid off, and ongoing rewards help cover operational expenses (OPEX), such as data center and internet service provider fees. For more detailed information, please refer to this wiki page.
The Gen-1 Remuneration Model currently published on the wiki can serve as a baseline. These rewards were calculated based on the need to cover the costs involved in setting up the equipment. With ongoing operating costs increasing annually, along with additional expenses it is proposed that Gen-1 NPs operate a maximum of 42 nodes while being compensated for operating costs.
Currently, operating costs account for approximately 47% of the total cost over 4 years. Data center costs have also risen significantly due to inflation and competition, with costs increasing by 25% over four years, or 6.25% annually. The annual operating cost of these servers is estimated to continue rising. Based on the information above, it is proposed that rewards be reduced by 33% from current levels after the initial 4-year period, and that this level of remuneration would remain in place for at least 2 years. This reduction would support the continued operation of Gen-1 NPs managing up to 42 nodes. To illustrate the approach, an example calculation is provided below for a 70-node NP based in Europe (excluding Switzerland):
Current node count: 70
Node count after 4 years: 42
Current remuneration per node: 1584 XDR
Reduction factor: 33% (0.67)
New remuneration per node after 48 months: 1584 * 0.67 = 1061 XDR
Total monthly rewards: 42 * 1061 = 44,562 XDR
In practice, the following remuneration would apply for Gen-1 nodes that have operated for more than 4 years:
TYPE-1 | Reward per node per month (in XDR) |
---|---|
US - California | 1072 |
US - other | 1004 |
Canada | 1088 |
Slovenia | 1152 |
Switzerland | 1136 |
EU - other | 1061 |
Singapore | 1234 |
Japan | 1188 |
Improving Decentralization Without Decreasing Total Node Count
In line with the target topology and the approved proposal #127044, the maximum number of nodes per NP will be set to 42. As a result, all NPs with more than 42 nodes will need to scale down, resulting in 308 excess nodes. Unlike proposal #127044, which mandates the complete removal of these nodes, we propose two options:
- Excess nodes may be sold to another NP while keeping the nodes in the same data center.
- Excess nodes may be sold to an optimal NP (one with few or no nodes), with the nodes moved to a new country and data center.
According to DFINITY’s analysis, if all NPs affected by the 42-node cap choose option 1, 30 subnets could be added, and IC node utilization would increase to 72% (up from 40%). If option 2 is chosen, 49 subnets could be added, and node utilization would rise to 89%.
Based on this analysis, option 2 is better for the network; however, the associated costs and risks might make it infeasible for some new NPs. Therefore, option 1 is deemed acceptable and is the mandatory action for all NPs with more than 42 nodes. Option 2 is preferred but not mandatory. The decision between either option is left to the respective NPs impacted by the node cap and those interested in taking on these excess Gen 1 nodes.
Please note that for the NP taking on the additional Gen 1 nodes, the remuneration will follow the terms outlined in this proposal, not the rates currently published for Gen 1 or Gen 2 nodes on the wiki.
Finally, the target topology prohibits onboarding new nodes. However, the new NPs taking on these excess nodes will be allowed to onboard only these specific nodes as this is considered a reassignment that maintains the total node count rather than increasing it by adding new nodes.
Why approved motion proposal #127044 needs to be updated
The current post-48-month rewards scheme proposal (proposal #127044) for Gen-1 NPs is such that:
- Operating 28-42 nodes is not profitable in top tier data centers in many of the current regions, which could lead to the entire operation being shut down rather than just scaled back.
- Any NP operating more than 42 nodes would need to shut down or sell the excess. This would lead to (at least) a 14% loss in active nodes or 18% in total nodes. Regaining this capacity to meet increased demand would be more expensive for the IC.
- A reduction in the number of active nodes, even by 14%, could trigger negative investor sentiment and potentially lead to a downward spiral in ICP price, discouraging both investors and developers from engaging with the platform.
Avoiding this reduction is in the community’s best interest for the following reasons:
1/ Be ready for the future:
We anticipate that the IC will see growth in the future beyond what was projected in this proposal.
The target topology currently shows that 40 subnets are needed to support current capacity. While Gen-1 nodes lack SEV-SNP, unlike Gen-2 nodes, they can still be used in 35 out of these 40 subnets. Moreover, most of these subnets (31 out of 40) are application subnets, which are the most likely to be added if demand grows. Maintaining the existing Gen-1 capacity allows the network to instantly accommodate sudden growth in demand and ensures new users receive the best possible experience during spikes in network activity.
Gen-1 NPs were brought onto the IC as they believed in the vision and future of the IC since before Genesis and have the scale, resources, and experience to quickly and reliably upgrade the IC’s infrastructure whenever required. These nodes were also placed in top tier data centers for increased resilience and security, both of which are still an asset to the IC These characteristics are not easily replaced. Forecasted below are the number of servers and the percentage of the network lost at various Gen-1 node caps.
Max Number of Nodes Kept by Gen-1 | Active Servers Lost | % of Network Lost | Total Servers Lost | % of Network Lost |
---|---|---|---|---|
56 | 55 | 4% | 112 | 7.5% |
42 | 189 | 14% | 266 | 18% |
28 | 357 | 26% | 440 | 30% |
14 | 583 | 42% | 678 | 46% |
0 | 835 | 60% | 936 | 63% |
Table 1: percentage of network lost at various Gen-1 node caps |
Please note that the number of servers lost may be higher as not all Gen-1 NPs have been identified and included in the figures of Table 1.
As noted before, proposal #127044 set a limit on the number of servers at 42 and adjusted the remuneration scheme. As operating costs for the nodes have increased dramatically since the IC’s genesis, many NPs may probably decide to stop running all their servers entirely, thus causing 60% of the network capacity to be lost. This would result in the IC no longer being at the target topology and could potentially destabilize the network.
2/ Keep sentiment high:
As explained above, even in the best case, there will be at least a 14% loss in the network. Many investors and analysts track the number of active nodes, so such a reduction by NPs to comply could be seen as a network failure, potentially leading to a downward spiral in sentiment and ICP price. A 14%-18% decrease in the network would be viewed unfavorably by investors. If some NPs decide to stop their entire operations, it would lead to an even larger decrease, further worsening sentiment. A similar reasoning applies to developers building applications that would drive further adoption. They may reconsider building on the IC if they see that the underlying infrastructure is vulnerable to many NPs pulling out.
3/ Cost-Effective Gen-1 Nodes
Onboarding new servers and NPs to address a potential shortfall in supply or sudden spike in demand is a time-consuming process. Ordering hardware, finalizing contracts with data centers and ISPs, and installing and deploying the nodes can take several months. This can cause new users to become frustrated with degraded service levels as they wait for the IC’s capacity to increase. Furthermore, this will also increase the inflation costs of ICP as the rising costs of new servers must also be compensated for.
In comparison, Gen-1 nodes are now more cost-effective for the network to run as the all initial purchase and setup costs have already been paid off. With this proposal, the average cost per Gen-1 node in key geographic areas would be 1117.06 XDR. In a 1 rack setup for Gen-2, the average cost per node is 1813.59 XDR, 62% higher than the current proposed new Gen-1 rate.
The outlined reasons emphasize the importance of avoiding node reduction, which is inevitable if the current post-48-month remuneration scheme outlined in proposal #127044 persists. To maintain current capacity, we propose a simple, efficient remuneration structure enabling all Gen-1 NPs to operate profitably.
Balancing Inflation and NP Incentives
Token Inflation:
Having followed discussions on the forum, we recognize the concerns regarding the contribution of NP rewards to token inflation. As the proposed updates in this proposal will alter rewards slightly, we would like to address this topic in this post as well.
Voting reward inflation is fixed and independent of the token price, whereas node provider-induced inflation is inversely proportional to the token price. With the appreciation of ICP over the last six months, the share of NP-induced inflation has significantly decreased. Let’s have a look at how these mechanisms work.
Inflation on the Internet Computer is driven by node rewards and voting rewards, as ICP is minted upon distribution. Staking rewards (maturity) are not minted unless they’re disbursed. For this reason, we only include disbursed voting rewards in our comparison. You can see the historical reward distribution of node rewards and disbursed voting rewards below:
The first node rewards were minted in June 2021, ever since the minting of node rewards takes place around the 14th of each month. The amount each node receives is converted to ICP from a fixed sum in XDR (special drawing rights); this allocation of XDR is determined based on the node’s location, hardware generation, and decentralization factors (for Gen-2). This means the total node remuneration in XDR is fixed to each node and, thus irrespective of the ICP price. However, what is affected by the ICP price is the amount of ICP a node provider receives. The monthly node rewards are converted into ICP based on a moving average of the ICP to XDR conversion rate. This ensures that NPs have a rather stable income to account for business expenses if they convert the received ICP into fiat currency. Do note that the moving average of the ICP to XDR conversion rate introduces a significant difference in conversion rate at the time of transaction time versus the allocated rewards based on this moving average.
Below is an overview of how ICP token price influences the node rewards distributed in ICP. Introducing more inflation by node rewards when the ICP token price is lower and the opposite when the price is higher. Again, note that this is derived from a fixed sum of XDR, a basket of fiat currencies, converted to ICP by following a moving average from the prior month.
Inflation by Node Rewards per Token Price:
ICP Token Price | Node Rewards (ICP/month) | AVG Disbursed Voting Rewards (ICP/month) | Relative Node Rewards Inflation Share | Total Yearly Inflation by Node Rewards |
---|---|---|---|---|
$2.50 | 1,273,292.45 | 1,042,895 | 54.97% | 2.94% |
$5.00 | 636,646.23 | 1,042,895 | 37.91% | 1.47% |
$10 | 318,323.11 | 1,042,895 | 23.39% | 0.74% |
$20 | 159,161.56 | 1,042,895 | 13.24% | 0.37% |
$50 | 63,664.62 | 1,042,895 | 5.75% | 0.15% |
$100 | 31,832.31 | 1,042,895 | 2.96% | 0.07% |
To get a full picture of inflation driven by ICP token price, we graph simulated node rewards and disbursed voting rewards at different price points. The following assumptions were made:
- Node rewards are a fixed number of active but stationary nodes, not moving location or transferring ownership.
- The node rewards are derived from a fixed allocation of 2,375,545.62 XDR per month, approximating today’s situation (June 2024).
- Voting reward disbursement does not fluctuate relative to the token price.
- The number for disbursed voting rewards is an average of disbursements within 1 standard deviation since July 2021.
- XDR to USD conversion rate changes are not taken into account.
We found that below $2.50 per ICP, the node rewards are the main driver of inflation on the network, representing more than 50% of inflation. Any price above results in a reduction of node rewards as an inflation driver. In the current model, node rewards represent 25% of inflation at a token price of $9.15 and drop further to 5% at $58.00 per ICP.
Even if current incentives for Gen-1 nodes are retained, inflation levels would be considerably lower than six months ago. However, under proposal #127044, many Gen-1 NPs would struggle to operate profitably and might cease operations. Additionally, service quality could deteriorate for those who persist as rewards fail to cover maintenance costs due to decreasing profitability.
Conclusion
In conclusion, avoiding the downsizing that is imposed in proposal #127044 for Gen-1 NPs is paramount for the community’s long-term success. This downsizing and potentially complete exit of Gen-1 Nodes could jeopardize the IC’s readiness for reaching product market fit.
Moreover, a reduction in active nodes, even by 14%, could trigger a downward spiral in ICP price, dissuading both investors and developers. A significant reduction in Gen-1 nodes would compromise the IC’s ability to efficiently serve high-demand regions, potentially leading to user dissatisfaction and increased costs for onboarding new servers.
Finally, since Gen-1 nodes are now more cost-effective for the network and can still be used in 35 out of 40 subnets, it’s crucial to maintain their operation. This helps prevent network capacity loss and avoids the inflationary costs tied to onboarding new NPs.
Therefore, we ask the community to vote to accept the proposed updates mentioned above. More concretely, voting to accept would mean that:
- The rewards for Gen-1 nodes operating beyond the initial 4-year period would be reduced by 33% from current levels, rather than the 75% reduction proposed in proposal #127044.
- This revised reward scheme would be in place for 2 years, providing Gen-1 NPs with the stability needed to make necessary commitments to their service providers, such as data centers, ISPs, and other necessary costs.
- NPs will be limited to operating a maximum of 42 nodes, in line with the target topology.
- Any excess nodes from the 42-node cap can remain on the IC. While reassignment to a “New” NP is mandatory, these nodes may continue in the same or a different data center and/or country. A “New” NP can also be an existing NP with fewer than 42 nodes. No NP will receive rewards for more than 42 nodes, and the new NP will receive the same 33% reduced remuneration for the acquired nodes. New NPs not yet listed in the NNS will be accepted through the self-declaration process and respective proposals.
- New NPs taking on these excess nodes will be allowed to onboard them without rejection due to target topology restrictions, as this is not an addition of new nodes but a reassignment that maintains the total node count and enhances network decentralization and utilization.
Thank you to all NPs and DFINITY team members for your contributions over the past months, from the Node Provider ICP Lab in May to now, bringing this discussion to a point where it’s ready to be shared with the community.
Thank you to the community for taking the time to review this proposal. We look forward to your feedback and questions in the forum thread below.
The goal is to submit a motion proposal exactly one week from now.
Please note that I am posting this in my capacity as the community lead of the Node Provider Working Group. This is separate from my responsibilities at Aviate Labs, founded as a division of Allusion, which is also a Gen-1 Node Provider. Allusion will make its own decision on how to vote on this proposal. My role here is purely to communicate the information.