1 of 17

DBC Computing Network

Overall system framework

Infinitely scalable network layer

Application layer for GPU virtual machine/container management

2 of 17

—— Overall system framework

Provides HTTP interfaces for querying node machine configuration information, creating GPU virtual machines, querying virtual machine configuration information, etc. Users can operate the virtual machines on a node through the client-end.

Customer-end Nodes

Pre-defined network nodes to provide alternate access portals for computing power network nodes.

Seed Nodes

Responsible for the management of GPU VMs/containers (e.g. creation, configuration information query, etc.), interaction with authentication servers (including: user identity verification, lease term check, etc.).

GPU Function Nodes

3 of 17

——GPU function node basic architecture and components

​

Linux（Host）

GPU

Hypervisor: KVM

​

​

GPU

GPU

GPU

.......

Network

HttpService

NodeService

  • Node Service: Responsible for GPU virtual �machine and container related operations

Mainly responsible for GPU virtual machine and container related operations, such as: creation, startup, shutdown, image management, lease status check, etc.

​

  • Http Service: Provide Http access interface to users externally

Users can access the GPU virtual machines in the computing network through the http interface for daily management.�

  • Network: Decentralized network service

Mainly responsible for joining and exiting nodes, as well as connecting and routing with other network nodes.

GPU

GPU

GPU

GPU

.......

windows(VM)

linux (VM)

4 of 17

—— GPU function node basic architecture and components

​

Linux（Host）

GPU

Docker

​

​

GPU

GPU

GPU

.......

Network

HttpService

NodeService

GPU

GPU

GPU

GPU

.......

Container

Container

  • Node Service: Responsible for GPU virtual �machine and container related operations

Mainly responsible for GPU virtual machine and container related operations, such as: creation, startup, shutdown, image management, lease status check, etc.

​

  • Http Service: Provide Http access interface to users externally

Users can access the GPU virtual machines in the computing network through the http interface for daily management.�

  • Network: Decentralized network service

Mainly responsible for joining and exiting nodes, as well as connecting and routing with other network nodes.

5 of 17

—— Overall Advantages

Flexible and free scalability

Machines from all over the world can simply and easily access the entire computing network, with completely decentralized management.

​

Massive amount of machine models

Ultra-low cost

Flexible and easy to use

A wide range of GPU models and configurations from around the world are available for users to choose from, and as the number of users joining increases, so does the number of machines available.

For users who need a lot of high-performance computing, saves the high cost of purchasing hardware and experience ultra-high performance computing equipment at a very low rental cost.

Both virtual machines and containers are supported, so users can rent them according to their own needs. The rental time is billed in seconds, hours and days, so you can rent as you go.

6 of 17

—— Overall System Deployment

User

HTTP

Client-end Node

Users can deploy multiple client-end nodes by themselves. Avoiding the failure of a single node, which could affect users' access to the computing network.

Client-end Node

Client-end Node

Seed Node

Seed Node

GPU Function Node

7 of 17

— Infinitely scalable network layer

​

Anyone can connect their machine to the Congtu Cloud Computing Network. When a new node starts, it accesses the entire computing network by connecting to the already existing computing network nodes. Each computing node periodically exchanges its respective neighbor nodes with neighboring nodes, and each node maintains a certain number of connections to neighboring nodes.

Connecting to the computing network

Exiting the computing network

When a node shuts down, the node adjacent to it removes the node from the list of neighboring nodes and exchanges the list of neighboring nodes with other nodes, and after a period of time, the shutdown node is removed from the entire network.

8 of 17

——GPU function node connecting to the network

P

N

A

B

peers： P B

peers： P A

peers： A B

Connect to existing nodes

peers：P

P

N

A

B

peers： P B

peers： P A

peers： A B N

Add to peers list

Add to peers list

P

N

A

B

peers： P B N

peers： P A N

peers： A B N

peers：P A B

Exchange peers list

Connect peers list

P

N

A

B

peers： P B N

peers： P A N

peers： A B N

peers：P A B

9 of 17

——GPU function node exiting the network

Close connection

P

N

A

B

peers： P B N

peers： P A N

peers： A B N

peers：P A B

Exchange peers list

P

N

A

B

peers： P B N

peers： P A

peers： A B

peers：P A B

Remove node N

Remove node N

P

N

A

B

peers： P B

peers： P A

peers： A B

peers：P A B

Remove node N

Remove node N

10 of 17

—— Client-end requesting a GPU function node

Client-end node

Request Message

Target node

11 of 17

—— Data transmission security at the network layer

​

  • To ensure absolute security of information transmission between nodes, asymmetric encryption is used to transfer messages between nodes.
  • Each node has a pair of public/private keys, and the public key is public within the computing network. Using the public key of the target node for encryption can effectively protect sensitive information transmitted between nodes (such as wallet address, GPU VM details, session_id, etc.).

node

node

Public�key

Private key

Public key

Private key

12 of 17

——GPU function node basic architecture and components

Encrypt with Node B’s �public key

Decryption failed

A

pub_key

private_key

B

pub_key

private_key

Client

pub_key

private_key

Decryption successful

Encrypt with Client’s �public key

Decryption failed

A

pub_key

private_key

B

pub_key

private_key

Client

pub_key

private_key

Decryption successful

[Client] sends a message to [Node B]

[Node B] sends a message to [Client]

13 of 17

Prevent messages from being broadcast indefinitely within the computing network

  • When a node broadcasts a message outward, each outgoing request carries a unique random number nonce, and each node will check and cache the nonce at the same time after receiving a request, and then broadcast the request
  • The next time a request carrying the nonce is received, it will be discarded after checking the cache and finding that it already exists, and will not continue to broadcast 。

node

node

node

nonce_check_ok

node

nonce_check_ok

nonce_check_ok

nonce_check_failed

(will not continue to broadcast the message)

Broadcasting for 1st time

Broadcasting for 2nd time

nonce

nonce

nonce

nonce

14 of 17

——Application layer for GPU virtual machine/container management

​

GPU VM/container

Basic operation

GPU virtual machine/container creation, shutdown, startup, mirroring, etc.

GPU VM/container

Status management

Unverified state, verified and online, leased state;

Periodically query the validation server to check the rental status of the virtual machine/container, the rental expiration will be retained for 10 days, and the virtual machine/container will be deleted again after 10 days of non-renewal of the rental.

GPU VM/container

Message monitoring

GPU virtual machine/container creation, shutdown, startup, mirroring and other operations

15 of 17

—— Verifying and renting machine

GPU computing node

Record GPU VM/container rental status

Record user wallet address, rent, rental period

Verifying server

Query rental status, check user rights

Monitoring �system

User

Creation, startup, shutdown, information query, mirror management

GPU function node

Regularly checks VM status

GPU VM/container

GPU VM/container

GPU VM/container

  • After a machine is launched on chain, it is in an unverified state at this time, and the verifier grabs the order to verify the configuration of that machine;
  • A machine can only have 3 validators and each has 3 time slots to validate the machine;
  • 3 validators check the machine for anomalies and check the configuration information by creating a virtual machine;
  • The machine becomes live when the validation is over, at which point it can be rented;

Verifying machine

  • User pays rent;
  • Creation of GPU VM/container by the user using the wallet address;
  • The renter can create and operate the GPU VM/container using their own wallet address or can also use a multi-signature account address; the request carries the wallet private key signature to ensure that the request comes from that user.

Renting machine

16 of 17

—— Rental code display

User/renter

Wallet address rental

Use multiple signatures

Account Address Rental

17 of 17

Building the world’s biggest distributed computing network.

Thank you!