Axion One：一个用 Rust 编写的神经符号微内核原型（关于调度器的帮助）
1 分•作者: yigremT•6 个月前
嗨 HN，
我是 AXION One 的创始人。我们正在使用 Rust (Stable 1.78) 构建一个神经符号微内核，尝试用向量符号架构 (VSA) 和联合嵌入预测架构 (JEPA) 来取代启发式资源分配。
问题：
在 Linux 上运行现代 AI 工作负载时，策略决策（调度、内存安全、能源配置）都留给了不透明的传统子系统。你实际上有一个黑盒（模型）运行在另一个黑盒（内核）之上，用户空间的基础设施试图弥合两者之间的差距。我们认为，主权 AI 需要内核本身成为一个能够持续学习和可验证推理的智能体。
架构（“如何做”）：
* VSA（向量符号架构）：我们将系统调用编码为高维全息向量 (D=10,000)。这使得内核能够使用恒定时间代数运算（绑定/捆绑）而不是复杂的分支逻辑来执行符号推理（例如，能力验证）。这为 IPC 提供了“可验证的粘合剂”。
* JEPA 调度器：我们没有使用 CFS，而是使用一个基于系统调用跟踪训练的小型自监督 JEPA 模型来预测延迟分布并选择帕累托最优线程顺序。
* 离线整合：在空闲周期内，内核通过 SGD“重放”事件图，以优化其嵌入，有效地“做梦”以优化未来的性能。
当前状态（我发帖的“原因”）：
我们有一个 GCP 原型（链接如上），验证了 VSA/JEPA 架构。它在仿真中通过了 93% 的 LKDTM 测试，并展示了自学习循环。
然而，树莓派 4 上的裸机 Rust 内核遇到了瓶颈。当我们 JEPA 调度器将更新后的嵌入写回内存时，我们看到由于缓存行争用导致 8–12 毫秒的内核停顿。我们还遇到了热恢复期间的竞态条件，导致 SDHCI 超时。
需求：
我正在寻找一位系统架构师联合创始人，他认为 8 毫秒的停顿不是失败，而是一个挑战。我需要一个理解以下内容的人：
* Rust nightly 和内核环境中的不安全边界。
* 无锁数据结构，以修复嵌入写入争用。
* 如何在重放缓冲区中实现因果掩码（目前缺失）。
我在这里回答关于为什么我们认为 VSA 是内核 IPC 的未来的问题。
AXION One 正站在系统编程重大突破的边缘。Rust、VSA 和 JEPA 的结合为 AI 基础设施中最紧迫的问题提供了理论解决方案：主权、可审计性和效率。然而，从经过验证的云原型过渡到强大的裸机内核需要特定水平的工程人才，即一位在复杂性中茁壮成长且不惧怕底层优化的系统架构师。
引用：
* Show HN: A DOS-like hobby OS in Rust.
* VEKOS: Verified Experimental Kernel OS.
* Show HN: Vect AI OS.
* Hacker News Posting Guidelines & Analysis.
* Vector Symbolic Architectures (VSA/HDC) Research.
* Joint Embedding Predictive Architecture (JEPA) Research.
* Redox OS & Rust Kernel Discussions.
查看原文
Hi HN,<p>I’m the founder of AXION One. We are building a neuro-symbolic microkernel in Rust (Stable 1.78) that attempts to replace heuristic resource allocation with a Vector Symbolic Architecture (VSA) and a Joint Embedding Predictive Architecture (JEPA).
The Problem:
Running modern AI workloads on Linux leaves policy decisions (scheduling, memory security, energy profiles) to opaque legacy subsystems. You effectively have a black box (the model) running on top of another black box (the kernel), with user-space scaffolding trying to bridge the gap. We argue that sovereign AI requires the kernel itself to be an agent capable of continual learning and verifiable reasoning.
The Architecture (The "How"):
• VSA (Vector Symbolic Architecture): We encode syscalls as high-dimensional holographic vectors (D=10,000). This allows the kernel to perform symbolic reasoning (e.g., capability verification) using constant-time algebraic operations (binding/bundling) rather than complex branching logic. This provides a "verifiable glue" for IPC.
• JEPA Scheduler: Instead of CFS, we use a tiny, self-supervised JEPA model trained on syscall traces to predict latency distributions and select Pareto-optimal thread orders.
• Offline Consolidation: During idle cycles, the kernel "replays" event graphs via SGD to refine its embeddings, effectively "dreaming" to optimize future performance.
Current Status (The "Why" I'm Posting):
We have a GCP prototype (linked above) that validates the VSA/JEPA architecture. It passes 93% of LKDTM tests in emulation and demonstrates the self-learning loop.
However, the bare-metal Rust kernel on Raspberry Pi 4 is hitting a wall. We are seeing 8–12ms core stalls due to cache-line contention when the JEPA scheduler writes updated embeddings back to memory. We also have race conditions during warm resume causing SDHCI timeouts.
The Ask:
I am looking for a System Architect Co-Founder who sees that 8ms stall not as a failure, but as a challenge. I need someone who understands:
• Rust nightly and unsafe boundaries in a kernel context.
• Lock-free data structures to fix the embedding write contention.
• How to implement causal masking in the replay buffer (currently missing).<p>I’m here to answer questions about why we think VSA is the future of kernel IPC<p>AXION One stands at the precipice of a major breakthrough in systems programming. The combination of Rust, VSA, and JEPA offers a theoretical solution to the most pressing problems in AI infrastructure: sovereignty, auditability, and efficiency. However, the transition from a verified cloud prototype to a robust bare machine kernel requires a specific caliber of engineering talent that a System Architect who thrives on complexity and is unafraid of low-level optimization.<p>Citations:
Show HN: A DOS-like hobby OS in Rust.
VEKOS: Verified Experimental Kernel OS.
Show HN: Vect AI OS.
Hacker News Posting Guidelines & Analysis.
Vector Symbolic Architectures (VSA/HDC) Research.
Joint Embedding Predictive Architecture (JEPA) Research.
Redox OS & Rust Kernel Discussions.