Introduction to GNET

Chapter 1. Introduction to GNET

1. 1. What is GNET

A biological neural network consists of neurons and synapses. A neuron is a living cell that exists from birth, responds to input signals, processes them, and transmits the result to other neurons. A synapse is a connection between neurons through which a signal is transmitted. The human brain has tens of billions of neurons, and orders of magnitude more synapses. From these two elements — and only these — all subjective reality is built: perception, thought, memory, will.

GNET (Gativus NETwork) is a specification for a digital network that reproduces this architecture. The Gativus network consists of two types of entities: nodes and relations. Nodes are digital analogs of neurons. Relations are digital analogs of synapses. Nothing else.

A node is a technical device that, like a neuron, has sensitivity to input signals, the ability to process them, and memory. Each node has a unique name by which it can be addressed from anywhere in the network. A node exists permanently from the moment of creation and cannot be deleted.

A relation is a directed transfer of a value from one node to another, similar to signal transmission through a synapse. A node computes a new value — it transmits it via a connector. The receiving node receives it and reacts. A neuron does not poll its neighbors — it fires and transmits a signal. Similarly, in Gativus, there are no requests and responses, no clients and servers. The only communication primitive is a datagram: one-way transfer of a typed value. There are nodes that live and exchange signals.

At the current technological stage, nodes are implemented as programs (operating system processes or other computing units). Relations between nodes are implemented over existing telecommunications networks.

GNET specifies the rules of this network: how nodes are named, how components inside nodes are addressed, how connectors are typed, how values are transferred, how new nodes are created, and how new relations are established. GNET is a platform-independent specification — it describes what should happen in the network without being tied to specific hardware. Implementation on a specific platform is described in a separate book — GATE.

1. 2. What is a node in Gativus?

A node is a digital entity located in the network that, like a neuron, has sensitivity to input signals, the ability to process them, and memory.

Each node has a unique name by which it can be addressed from anywhere in the network. A node exists permanently from the moment of creation and cannot be deleted. Like a neuron, it lives as long as the system lives.

Inside a node — components. A biological neuron contains dendrites (input), soma (processing), axon (output), and synaptic endings (connectors). A Gativus node contains analogous functional components, grouped into sections. Each component has a unique name within the node's local space.

At the current technological stage, nodes are implemented as programs, more precisely — as operating system processes or other computing units. Each node is a separate process (which has a name) with its own memory and executable code (resources allocated to the process).

The formal structure of a node is defined in Chapter 3.

1. 3. What is a relation in Gativus?

A relation is a directed transfer of a value from one node to another, similar to signal transmission through a synapse. Graphically, a relation is depicted as a directed line from the output of one node to the input of another — analogous to an axon–synapse–dendrite link.

A neuron does not poll its neighbors — it fires and transmits a signal. Similarly, in Gativus, there are no requests and responses, no clients and servers. There are nodes that live and exchange signals. The only communication primitive is a datagram: one-way transfer of a typed value.

Relations are established not between nodes as a whole, but between their components. For each relation, a component creates a connector — a named entry or exit point. One connector participates in one relation. A component can have many connectors — as many as it needs relations.

Each connector has a unique address inside the node. A relation is uniquely defined by a pair: source connector address and destination connector address. This pair is the relation identifier.

The datagram format and relation types are defined in Chapter 6.

1. 4. Technological prerequisites

Technological prerequisites

The GNET network is not an isolated infrastructure. It exists on top of the Internet, using existing technologies for transport and storage.

a) Internet

IPv6 is the transport foundation of the Gativus network. Node identification and component addressing are aligned with IPv6 standards and allow packet formation without additional encapsulation layers. Relations are virtualized over IP — akin to "Synapse over IP". To accelerate critical paths, GNET provides alternative routing methods within a single device and a single subnet.

b) Computing platform

Existing computing platforms (Intel, ARM, MIPS) and open-source operating systems (FreeBSD, Linux) are used to host nodes and perform IP routing. Such systems contain the necessary storage and routing components, although they will require kernel extensions to achieve the required performance and data integrity.

The specification of the edge device GATE, addressing GATN/GANN, and the routing protocol GORP are defined in Chapter 4.

1. 5. Network governance, creation, and security

Gativus builds an approach alternative to traditional programming and deployment. Instead of writing programs and initializing them — an ecosystem of a self-deploying morphogenetic network is created.

To this end, a hierarchy of rights issuers is established: the root registry issues licenses for edge devices, containing quotas for creating nodes and relations. Licenses are divided and delegated down the tree — from the root registry to the operator, from the operator to administrative nodes, from administrative nodes to worker nodes. Each act of creating a node or relation consumes a quota. The total volume of resources in the system is invariant.

Security is ensured by a special component of each node: a rights matrix determines which operations are allowed for each class of partners; relation encryption is performed via key exchange when establishing a connection.

Economics and topology — Chapter 4. Morphology — Chapter 5. Security — Chapter 7.