sub-modules
Performance Substrate
Vendor benchmarks lie by omission — ours ship with the receipts attached.
Every infrastructure pitch shows a beautiful benchmark and hides the conditions that produced it. Performance Substrate inverts the ritual: tiered compute from baseline to accelerators, each run carrying its execution profile — device, inputs, outcome. Claims arrive with the receipts to check them, on your hardware, under your load.
Substrate tiers
Each tier below builds on the last — workloads climb only as far as their evidence and deployment require.
data Viz
data View
section: stages
Baseline and Concurrency
Tier 0 baseline execution with Tier 1 concurrent scaling across customer workloads. Tenancy modes compile per deployment so shared and dedicated runtimes get exactly the isolation they promise.
Accelerators and Storage
Accelerator families attach per workload with execution profiles recording device, inputs, and outcome. The storage trio serves transactional, analytical, and cache needs under explicit contracts.
Streaming and Retrieval
Event streaming carries canonical envelopes with replayable positions. Vector retrieval profiles serve semantic search with freshness guarantees instead of stale indexes.
Industrial Gateway
The native sensor and SCADA gateway speaks open protocols — IEC, ISO, Modbus, MQTT, MTConnect — built against published standards. Undocumented or restrictively licensed protocols are refused by policy with the reason named.
Substrate guarantees
- Profiled runs
- Execution profiles with device, inputs, and outcome per accelerated workload
- Open protocols only
- Adapters require published standards or documentation; refusals name the reason
- Topology-matched tenancy
- Isolation modes compiled to fit the deployment, from SaaS to air-gap
- section
- capabilities
Substrate by deployment
Performance posture bends per deployment — same tiered model, different hardware and isolation stakes.
data Viz
data View
section: industries
Cloud and SaaS
Elastic scaling with accelerator pools shared under tenancy modes. Bursts provision against cost guardrails with the spend attributed per workload.
Self-hosted
Customer hardware with topology-matched tenancy compiled in. Capacity plans from measured profiles instead of vendor sizing guides.
Edge and Air-gap
Ruggedized tiers with local execution profiles and signed bundle updates. Sensor gateways stream through store-and-forward where links are intermittent.
Supported protocol matrix
- Modbus (Organization)
- Open specification — full adapter shipped against published standard
- MQTT and OPC UA
- Open standards (OASIS, IEC) — full adapters with pub/sub and information-model support
- MTConnect and ODVA
- Open institute and association specs — adapters built directly against standards
- Bluetooth SIG and KNX
- Association-published specs — adapters for sensor and building telemetry
- GS1 and EPCIS
- Open supply-chain standards — adapters for identity and custody events
- NMEA and IEEE families
- Association and institute standards — adapters for positioning and instrumentation
- Vendor REST and WebSocket APIs
- Shipped where vendors publish documentation freely without NDA
- Open-source libraries
- Adopted where licensing permits; GPLv3 and AGPL isolated out of core
- Undocumented protocols
- Reserved slot only — no production adapter loads or parses live traffic
- Research-blocked adapters
- Runtime denial with named reason instead of obscure failure
- section
- protocol-matrix
Performance Substrate questions
Section 1
How are performance claims verified?
Execution profiles record device, inputs, and outcomes per run on customer infrastructure. Claims cite reproducible benchmarks, never vendor-lab numbers alone.
Section 2
What if our machine protocol is proprietary?
The gateway refuses it by policy with the reason named — then offers the documented-adapter path: publish the protocol, provide an open library, or expose a REST API, and the adapter ships.
Section 3
Do accelerators mean replatforming risk?
No. Workloads climb tiers only as evidence requires, with tenancy and deployment topology matched. Baseline execution stays available as the fallback that always works.
data View
section: faqs
content section
value: Every tier carries its execution profile: what ran, where, on which inputs, with what result — performance claims arrive with the receipts to check them.
Section 7
Tiered scaling from baseline through concurrent to accelerated execution
Section 8
Compile-time tenancy modes matched to deployment topology
Section 9
Accelerator families with execution-profile taxonomy per run
Section 10
Storage trio tuned per workload with cache contracts
Section 11
Event streaming with replayable consumer positions
Section 12
Vector retrieval profiles for semantic search workloads
Section 13
SCADA and sensor gateway built on open protocol standards
Section 14
Deny-by-default adapter policy with named refusal reasons