Technology

Infrastructure-independent computing.

Softadastra is developing an adaptive computing architecture where software describes what it needs and the Runtime determines how available infrastructure can satisfy it.

Runtime source ↗ Research →

software Intent + State + Constraints

execution Softadastra Runtime in progress

resources Computing Fabric

Core architecture

Application

Describe requirements

Application code expresses intent, state requirements and constraints without permanently encoding a complete infrastructure topology.

intent · state · constraints

Runtime

Make execution decisions

The Softadastra Runtime interprets application requirements and progressively decides how computation and state should use available resources.

placement · scheduling · state · recovery

Fabric

Expose capabilities

Local machines, edge systems, cloud infrastructure and hardware accelerators become resources available to the Runtime.

local · edge · cloud · CPU · GPU

Software describes the requirement. The Runtime handles the infrastructure decision.

Softadastra Runtime

The adaptive execution layer between software and computing infrastructure.

The Runtime is currently in progress. It is being developed publicly and individual capabilities are being implemented and validated progressively.

github.com/softadastra/runtime ↗

Runtime area	Observes	Responsibility
Placement	latency · locality · trust · capabilities	Determine where computation should execute across available infrastructure.
Scheduling	priority · load · resources · constraints	Schedule workloads according to application requirements and current resource conditions.
Distributed state	durability · consistency · availability	Manage application state independently from any single machine or location.
Replication	state · availability · locality	Determine when and where state should be replicated across the computing fabric.
Recovery	failure · availability · application state	Recover or resume execution when machines, networks or infrastructure become unavailable.
Reconciliation	divergent state · connectivity · consistency	Reconcile state after partitions, offline operation or infrastructure reconnection.
Trust	identity · policy · security	Restrict execution and state placement to infrastructure permitted by application policy.
Hardware	CPU · GPU · accelerators · capabilities	Select compatible computing resources without permanently coupling the application to specific hardware.
Adaptation	network · resources · failures · changing conditions	Re-evaluate execution decisions while the application and its environment change.

Computing Fabric

The Fabric is the set of computing capabilities visible to the Runtime.

It is not another cloud provider. A resource may be local, remote, temporary or specialized.

The application should not need to permanently depend on the location or provider of that resource.

control Softadastra Runtime

local Local machine CPU · memory · storage

edge Edge resource locality · low latency

cloud Cloud resource remote capacity

acceleration GPU / accelerator specialized compute

connected local + remote resources

network unavailable eligible local work continues

reconnected state can be reconciled

Technology map

maintained

Vix.cpp

Developer platform

Open-source C++ developer platform maintained by Softadastra.

open ↗

in progress

Softadastra Runtime

Adaptive execution layer

The central runtime technology behind infrastructure-independent computing.

open ↗

planned

Softadastra Node

Reference hardware

Low-power reference hardware designed to run the Runtime as one resource in the computing fabric.

details ↓

Softadastra Node planned

Reference hardware for proving the Runtime on a small, ordinary, low-power machine.

The Node is not required to use Softadastra. It is one future computing resource designed specifically around the Runtime.

The first target intentionally favors low cost and low power over workstation-class hardware.

	node v1 reference target
Role	Reference hardware for the Softadastra computing fabric
Architecture	x86-64 initially
Prototype class	Low-power 4-core system
Memory	16 GB target
Storage	512 GB NVMe target
Networking	2.5 GbE · Wi-Fi
Power	< 20 W typical system target
Operation	Local-first · able to remain useful without Internet connectivity
Status	Planned reference system

Target specification. Final hardware may change as prototypes and the Runtime are validated.

The architecture is being built progressively.