Slice pricing and bid system #21

New issue

Open

opened 2026-01-12 22:12:42 +00:00 by scott · 0 comments

scott commented

2026-01-12 22:12:42 +00:00

Member

We want to implement a pricing/bid system modeled after vast.ai. Hosters should be able to set different prices for the different modes of reservation.

AI Summary of the vast.ai system

Comparison Summary

Tier	Industry Name	Cost	Reliability	Commitment
On-Demand	Guaranteed / Standard	High	High	None (Pay-as-you-go)
Interruptible	Spot / Preemptible	Low	Low	None (Can be kicked off)
Reserved	Committed / Prepaid	Medium-Low	High	Long-term (1 month to 3 years)

1. The "Guaranteed" Tier (On-Demand)

This is the baseline for professional computing. You pay the standard market rate to ensure that the resource is yours for as long as you need it.

The Logic: You are paying for availability and peace of mind. The provider agrees not to let anyone else use that hardware while you are paying for it.
Reliability: 100% (or as close as the provider's SLA allows). It will not be shut down unless you stop it or fail to pay.
Pricing: Fixed and usually the most expensive per hour.
Best For: Critical "always-on" services, customer-facing apps, or time-sensitive deadlines where a restart would be catastrophic.

2. The "Spare Capacity" Tier (Interruptible / Spot)

Cloud providers often have "idle" hardware that isn't currently being rented at the full On-Demand price. Rather than let it sit dark, they rent it out at a massive discount (often 60–90% off).

The Logic: You are paying for leftovers. In exchange for the low price, you agree that if a "Guaranteed" customer comes along and needs that hardware, the provider can kick you off with little to no warning.
The Bidding Aspect: In many marketplaces, if multiple people want the same "leftover" resource, it goes to the highest bidder. If someone bids higher than you, your job is paused or terminated.
Reliability: Low. Your work can be interrupted at any time.
Best For: "Batch" tasks that can be easily resumed, such as rendering video frames, training AI models with frequent checkpoints, or large-scale data processing.

3. The "Committed" Tier (Reserved / Prepaid)

This is a hybrid model designed for users who know exactly how much power they need for a long period (months or years).

The Logic: You are paying for loyalty and predictability. By telling the provider, "I promise to use and pay for this machine for the next 12 months," they give you a discount similar to the Interruptible tier, but with the reliability of the Guaranteed tier.
Reliability: 100%. Since you’ve already paid (or committed to pay), the resource is locked to your account.
Pricing: Deeply discounted, but usually requires an upfront payment or a legal contract.
Best For: Established businesses with stable, predictable traffic or long-term research projects.

How to choose?

If the job must finish by 5:00 PM today: Use On-Demand.
If the job is long and your budget is tiny (and you don't mind it taking an extra day): Use Interruptible.
If you are running a server that will be active for the next year: Use Reserved.

We want to implement a pricing/bid system modeled after vast.ai. Hosters should be able to set different prices for the different modes of reservation. ## AI Summary of the vast.ai system ### Comparison Summary | Tier | Industry Name | Cost | Reliability | Commitment | | :--- | :--- | :--- | :--- | :--- | | **On-Demand** | Guaranteed / Standard | High | High | None (Pay-as-you-go) | | **Interruptible** | Spot / Preemptible | Low | Low | None (Can be kicked off) | | **Reserved** | Committed / Prepaid | Medium-Low | High | Long-term (1 month to 3 years) | ### 1. The "Guaranteed" Tier (On-Demand) This is the baseline for professional computing. You pay the standard market rate to ensure that the resource is yours for as long as you need it. * **The Logic:** You are paying for **availability and peace of mind**. The provider agrees not to let anyone else use that hardware while you are paying for it. * **Reliability:** 100% (or as close as the provider's SLA allows). It will not be shut down unless you stop it or fail to pay. * **Pricing:** Fixed and usually the most expensive per hour. * **Best For:** Critical "always-on" services, customer-facing apps, or time-sensitive deadlines where a restart would be catastrophic. ### 2. The "Spare Capacity" Tier (Interruptible / Spot) Cloud providers often have "idle" hardware that isn't currently being rented at the full On-Demand price. Rather than let it sit dark, they rent it out at a massive discount (often 60–90% off). * **The Logic:** You are paying for **leftovers**. In exchange for the low price, you agree that if a "Guaranteed" customer comes along and needs that hardware, the provider can kick you off with little to no warning. * **The Bidding Aspect:** In many marketplaces, if multiple people want the same "leftover" resource, it goes to the highest bidder. If someone bids higher than you, your job is paused or terminated. * **Reliability:** Low. Your work can be interrupted at any time. * **Best For:** "Batch" tasks that can be easily resumed, such as rendering video frames, training AI models with frequent checkpoints, or large-scale data processing. ### 3. The "Committed" Tier (Reserved / Prepaid) This is a hybrid model designed for users who know exactly how much power they need for a long period (months or years). * **The Logic:** You are paying for **loyalty and predictability**. By telling the provider, "I promise to use and pay for this machine for the next 12 months," they give you a discount similar to the Interruptible tier, but with the reliability of the Guaranteed tier. * **Reliability:** 100%. Since you’ve already paid (or committed to pay), the resource is locked to your account. * **Pricing:** Deeply discounted, but usually requires an upfront payment or a legal contract. * **Best For:** Established businesses with stable, predictable traffic or long-term research projects. --- ### How to choose? 1. **If the job must finish by 5:00 PM today:** Use **On-Demand**. 2. **If the job is long and your budget is tiny (and you don't mind it taking an extra day):** Use **Interruptible**. 3. **If you are running a server that will be active for the next year:** Use **Reserved**.

No labels

urgent

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

lhumina_code/hero_ledger#21

No description provided.

Rows
Columns