What question did this study set out to answer?

The aim is to showcase a data-only execution system that enhances performance and security without traditional compilation processes.

February 19, 20260 citationsOpen Access

Implementing a Tall-Infra Data-Only Execution System

GHGeoffrey Howland

Key Points

The aim is to showcase a data-only execution system that enhances performance and security without traditional compilation processes.
Developed Silo as a data-only execution system.
Utilized a universal Entity container and Prolog logic for behavioral management.
Employed multiplicative Utility AI scoring and stack-based Logic Blocks.
Achieved network security through fixed-size packets and access control.
Implemented NUMA-aware work distribution for optimal CPU scaling.
Maintained 60 FPS real-time performance.
Attained 95%+ CPU utilization per core.
Updated 10,000 entities in 4-8ms per frame.
Achieved zero iteration time for immediate changes.
Ensured geometric security against privilege escalation and data exfiltration.

Abstract

Implementing a Tall-Infra Data-Only Execution System Using the Silo Engine as an Example We present Silo, a data-only execution system that eliminates compilation from the development loop while maintaining 60 FPS real-time performance and perfect CPU scaling. The architecture uses a universal Entity container, declarative Prolog logic, multiplicative Utility AI scoring, and stack-based Logic Blocks to create a system where all behavior resides in hot-swappable data tables rather than compiled code. We demonstrate geometric security through fixed-size network packets and path-based access control, achieving input isolation without sandboxing. The system scales linearly to 64+ cores via NUMA-aware work distribution with barrier synchronization. Complete reference implementations of all core structures (Entity, StateMachine, Prolog, UtilityAI, LogicBlock, Thread, Networking) are provided. Measured results: 10,000 entities updated at 4-8ms per frame, 95%+ CPU utilization per core, zero iteration time (changes take effect next frame), and architectural impossibility of privilege escalation or data exfiltration from network inputs. Package Contents:* `manuscript.md`: Paper* `README.md`: Overview **Status:** Architectural Blueprint for Independent Implementation

AIに質問

Bookmark

View Full Paper