People have been trying to come up with new ways to combine blockchain technologies with existing concepts and services since 2017. One of these ideas was blockchain-based cloud computing. However, the technology wasn’t ready for this kind of application at that stage. Many projects have tried and failed — Sonm, iExec, Hypersphere, Golem, all of them are either dead or barely visible now. Some projects, such as the Render Network, still operate pretty successfully, having their own clients and a large number of nodes. Overall though, the potential of the cloud computing market is still untapped by blockchain technologies. Well, here comes CryptoRat — powered by Everscale.
What is CryptoRat?
CryptoRat is a platform for decentralized deployment of AI models, designed for data scientists. The market for cloud AI computing is huge — almost $40 billion in 2020 according to Qualcomm, and projected to exceed $70 billion by 2025. The demand for hardware resources for machine learning doubles every quarter, and hardware manufacturers struggle to keep up with it. Yet the majority of existing hardware stays idle half of its lifetime, in data centres and in homes. It’s thus possible to put this to good use and give it from 30 to 50% additional workload by assigning these machine learning tasks to it. That’s exactly what CryptoRat tries to solve.
Why have so many similar projects failed on their path? The CryptoRat team, seasoned data scientists in their own right, analysed all their competitors and came to a conclusion: those who succeeded supported only one type of task on their platforms, while those who failed tried to introduce as many types of tasks as possible. CryptoRat therefore took the former approach, with the main focus being a hosting platform for already trained computer vision models. The whole distributed network works on the same basic task, and it’s possible to manage all incoming jobs from customers effectively.
A distributed network of workers
There are three type of actors in the network:
- Users providing computing resources (regular PCs with GPU, data centres)
- Customers willing to use these resources
- The main router
It’s similar to a BitTorrent network. The router takes ONNX-compatible models from B2B customers and saves them in storage. The B2B customer can send requests to the router, which gives them addresses of the worker nodes according to their computing power, and the customer works with these directly, sending tasks and receiving results. The nodes are always connected to the router, asking it if there’s any job available.
The interaction between the protocol and the customer is established via the SDK. This is written in Python and distributed under an open source license, but to use it, it’s necessary to get an API key from CryptoRat. All models use the same ONNX format, the industry standard for data scientists, supported by scikit-learn, PyTorch, and synaptic frameworks. Each model gets its own ID after being deployed, and the customer can use it by addressing the API to set tasks for worker nodes.
The technology has come a long way since 2017, and now it’s possible to create software that can run on any PC without any additional setup or requirements. CryptoRat offers the simplest interface for users willing to run a node — just an executable file, both for Windows and Linux, which enables them to join the network from the get-go. The team believes that the gap between Web2 and Web3 should be closed, and that end users shouldn’t have to be experts in blockchain or Python environments. For them, everything is similar to Web2, except for one thing: payments.
Payments on Everscale infrastructure
Every user receives rewards for their network participation in USD, issued on Everscale. There are three main factors which determine the payment rate:
- Uptime
- Computational power
- Reputation
The longer the user stays in the network and the more uptime they have, the more reputation points they get. In the future, reputation will be represented by special tokens, which can be bought and sold among users. If the user doesn’t behave properly, e.g.. leaves the network with unfinished tasks, their reputation stake will be slashed.
As a result, all the worker nodes connected to the network will receive payment for each hour of uptime and get reputation points for every three hours of uptime. If the node stays idle, it still gets payment, and its resources are still reserved by the protocol.
Where does the money come from?
All worker nodes get payment, and it doesn’t come from nowhere. The CryptoRat team is building a B2B project which targets traditional business, so the payment is received in fiat currencies and distributed among users based on their input. Customers have two options of payment:
- Payment per request, which suits for one-time tasks, demos, and temporary solutions
- Payment per instance — to have the model and its API always ready to use
Each model has its pros and cons, but in both cases, customers can work with the protocol without even knowing it’s blockchain-based, and that’s the easiest way to achieve mass adoption.
Selling points
Why should data scientists choose CryptoRat over other solutions, such as AWS? First of all, they can perform four to six times more tasks for the same amount of money, according to estimates..
The second, and probably even more important point is time saved on deploying models. Data scientists spend 50% of the time necessary to set up models for work on deploying them. CryptoRat simplifies all the infrastructure work by deploying the models automatically and guaranteeing that the API serving the model will be available anytime.
Benefits for Everscale
CryptoRat is a great showcase which will introduce Everscale to many traditional businesses and users. Users will need to create an Everscale wallet address to receive Ever USD rewards and exchange them for fiat or other crypto tokens on FlatQube. Further down the road, the whole protocol will become completely decentralized, using Everscale’s infrastructure. We therefore regard CryptoRat as a project with high potential and wish the team the best of luck!
