Resources
Mission: Ethernet for the Next 50 Years
Open Atomic Ethernet (OAE) exists to restore coherence, unity, and long-term stability to the Ethernet ecosystem. Over decades, Ethernet has evolved into a fragmented family of partially overlapping standards, proprietary extensions, and incompatible interpretations. This fragmentation slows innovation, raises costs, diminishes interoperability, and weakens the ability of the global community to rely on Ethernet as the universal communications substrate for modern computing.
OAE’s mission is to reunify these fragments, re-establish a clear technical foundation, and bring the next generation of Ethernet back under a coherent and forward-looking IEEE 802.x framework. Our goal is nothing less than a single, strong, open standard that supports reliability, determinism, atomic transactions, and modern workloads from cloud to chiplet fabrics.
OAE Project Core Principles
- Excellence through Cadence
- Radical Simplicity
- Unified semantics across all scales
- Open Governance and return to IEEE Standards Stewardship
Companion Paper
Open Atomic Ethernet: Recovering the Bilateral Primitive that the Last Fifty Years Forgot (PDF, 9 pp.)
Paul Borrill, DÆDÆLUS Inc. — Presented at the Asilomar Microcomputer Workshop #52, 6 May 2026 (Athematic Session, K. Shoemaker chair).
The paper shows that the original Ethernet (Metcalfe & Boggs, 1976) embedded a three-phase bilateral handshake — preamble, ENDREPLY, the inter-record gap — which made every transfer evidentially mutual. The bug arrived later, as the field optimised the forward channel only and built a discipline of impossibility results (FLP, CAP, Two Generals) on the unstated assumption that information must flow in one temporal direction. OAE recovers what the original Ethernet knew, and generalises it: every link is two simultaneous Shannon channels, every transaction is bilateral, every commitment is mutual.
What makes OAE different
Most contemporary link efforts — NVLink, UALink, Ultra Ethernet, TTPoE — extend existing engineering practice: faster lanes, tighter silicon, better software stacks layered on the same fire-and-forget primitive Ethernet has carried since 1976. Manoj Wadekar’s ESUN workstream at OCP names the resulting gap directly: today’s Ethernet semantics are not sufficient for AI-scale workloads. OAE proposes the bilateral primitive that closes it.
OAE begins one layer deeper than incremental optimisation. It starts from a mathematical structure: a bilateral primitive with named invariants — Evidence-before-commit, Fail-closed, Reflective observation, Bounded ambiguity — that a protocol either satisfies or does not. Because the foundation is a structure rather than a behaviour, everything that sits on top of it becomes easier to do well. The protocol is easier to reason about, easier to formally specify, easier to verify, easier to simulate, and — because the work the wire is doing matches the work the application actually needs — easier to make both fast and reliable. The performance characteristics (ultra-low latency, deterministic failure semantics) are consequences of the mathematics, not goals pursued against it.
Open Æthernet Specification (v0.8, draft)
Open Æthernet (OAE) — published through v0.7a as Open Atomic Ethernet.
From v0.8 the Specification is published as per-chapter documents — each chapter is an independently versioned PDF, and there is no single monolithic download. The v0.8 draft chapters (July 2026) are hosted here: spec-v0.8 folder on GitHub. The earlier complete spec remains available as one PDF: Open Atomic Ethernet — Specification v0.7a (241 pp.).
v0.8 chapters (draft — click a title to open the PDF):
- Front matter — Conformance & Normative Language · Terminology · Changelog
- Ch. 1 — Principles of Operation (informative)
- Ch. 2 — Bits and Bytes (normative — nSLICE encoding changed in v0.8, see changelog)
- Ch. 3 — The End-Dally and the SACK Ladder (normative, NEW — the canonical Shannon/Spekkens/Turing/Metcalfe rung naming)
- Ch. 4 — Cells and Links (normative — now includes the Cluzter: bounded set membership as graph theory)
- Ch. 5 — Addressing and Routing (normative)
- Ch. 6 — Reversible Transactions (normative — now includes Repoch: reversible epochs, no unbounded counters)
- Ch. 7 — Architecture (normative)
- Ch. 8 — The Slice Engine (SerDes) (normative — slice architecture, SAW protocol, PIF, TPI, formal specification)
- Ch. 9 — Link Transaction Manager (normative)
- Ch. 10 — Topology (informative)
- Ch. 11 — Ultra Ethernet, Context and Comparisons (informative)
- Ch. 12 — History (informative)
- Ch. 13 — Theory (informative — now includes Monotonicity, CRDTs, and the Missing Bilateral Commit)
- Ch. 14 — Requirements and Use Cases (informative)
- Annex A — A Formal Model of Perfect Information Feedback (normative — fault model F1–F5, safety/liveness split, theorems O1/O3 discharged)
v0.8 highlights: canonical SACK-rung naming; nSLICE now counts up with rung depth (one-line bit inversion migrates v0.7a implementations); Timeout-and-Retry removed from every normative automaton (timeouts survive only as liveness bounds inside the RLFD — expiry can declare failure, never success); normative PIF formal model with two discharged theorems; exactly-once claims corrected per P. S. Almeida (received vs. delivered distinction; Exon, ICCCN 2022).
Fast Intro
OAE begins from a mathematical structure — a bilateral primitive with named invariants — making the protocol easier to reason about, formally specify, verify, simulate, and make fast.
Open Atomic Ethernet — Recovering the Bilateral Primitive that the Last Fifty Years Forgot (Asilomar Microcomputer Workshop #52, 6 May 2026)
Ethernet’s evolution is being shaped by a transformative paradigm: the Chiplet Economy. This new era is defined by modular semiconductor designs where monolithic chips are replaced by smaller specialized silicon dies, fostering a ecosystem of interconnects, standards, and applications. This modularity enables heterogeneous integration of compute, memory, and I/O dies, accelerating innovation cycles on specialized system architectures. It also introduces new challenges, particularly in interconnectivity. Chiplets within the same package require low-latency, high-bandwidth communication, which existing on-chip interconnects like PCIe or CXL attempt to address. However, when chiplets span packages or systems, on-chip controller based protocols are unable to scale their guarantees, and Ethernet must step in to provide seamless, standardized communication.
The industry must rethink the relationship between networks and applications. Applications demand a network that guarantees certainty—where communication is reliable, scalable, and invisible until needed. This requires abandoning the outdated “best-effort” model of the past 50 years and replacing it with a system where messages behave like transactions: either both software endpoints confirm success, or failure leaves no trace.
This entanglement between sender and receiver, with no library, software, or hardware permitted to disturb that entanglement, is a fundamental change from the ”throw it over the wall and hope it gets there” of the last 50 years. From a different perspective, a network message becomes a transaction which exhibits ACID properties: atomicity, isolation, and durability directly, and consistency because that message exists only between endpoints and is not visible to any other observer. Æthernet seeks to:
- Rethink Ethernet’s core primitives: frame structure, error recovery, and flow control, in the context of Shannon Information Theory.
- Introduce API’s for programmable minimal compute capability at the NIC level to enable reversibility and support application-specific protocol behavior.
- Address the trade-offs between signals, noise and energy dissipation, building upon principles from Quantum Thermodynamics.
- Create a protocol framework that is as simple as Ethernet’s original design but scalable to today’s and tomorrow’s use cases, including chiplet interconnects and AI accelerators.
- Introduce new fundamental multi-phase primitives to instruction sets: Set Reconciliation of Shannon Slots for common knowledge, and Network assisted transactions.
- Enable host bypass, allow direct use of layer 1/2 networking, and support NIC extension via “applets” for tasks like in-network computation.
OAE in the Link Landscape
OAE is one of several contemporary efforts addressing the limits of current link standards. The differences are visible most clearly along a single axis: whether the effort proposes a new wire / better stack on the existing Forward-In-Time-Only (FITO) primitive, or proposes a new primitive that the wire and stack express.
| Effort | Scope | Relationship to OAE |
|---|---|---|
| NVLink (NVIDIA) | Proprietary, GPU-centric scale-up fabric | Different category: vendor-locked, single-purpose. OAE is open and not GPU-specific. |
| UALink | Consortium-driven open scale-up link | Same FITO primitive, new wire. OAE replaces the primitive. |
| Ultra Ethernet (UEC) | Ethernet-based scale-out for AI/HPC, performance-tier | Performance optimisation on FITO. OAE is a semantic re-foundation. |
| ESUN (Meta / OCP) | Workstream identifying that current Ethernet semantics are insufficient for AI clusters | ESUN names the gap; OAE proposes the bilateral primitive that closes it. |
| TTPoE (Tesla) | Transport-layer protocol for in-vehicle networks | Different layer. OAE operates at L2 with bilateral semantics. |
In short: where the others optimise an existing primitive, OAE replaces it — and ESUN’s diagnostic work helps the industry see why the replacement is needed.
The Link Wars: Is the debate over or is it just beginning? panel at Chiplet Summit 2026 (January 23, 2026, Santa Clara) brought industry leaders to one table to debate exactly this question. Paul Borrill organized, and Bill Wong moderated; participants discussed NVLink, UALink, Ultra Ethernet, ÆLink, and TTPoE. The one-page panel handout is here: Link Wars panel description (PDF).
2026 Conference & Standards Contributions
In July 2026 the Open Atomic Ethernet program was presented in the same week at two venues that, read together, bracket the whole stack. The IEEE 802 July Plenary (802.1 Nendica, Montréal, 12–17 July) established the bilateral-commit model at the bottom of the stack — the link layer, the bits on the wire. Netdev 0x1A (Rome, 13–16 July) established the same model at the top — the Linux system-call boundary between application and kernel. One consistent, verifiable model of committed transfer runs the whole height of the stack, from write() to SerDes: the atomicity lattice of the Montréal atomicity paper and the Kernel Acknowledgment Spectrum of the Rome “Why Syscalls Fail” paper are the same ladder, measured at two boundaries.
IEEE 802.1 Nendica — Montréal Plenary (12–17 July 2026)
Two contributions were presented to IEEE 802.1 Nendica (the ICne activity, chair Roger Marks). Both are public on IEEE Mentor; the citable permanent identifier for each is its DCN. All URLs and revisions were verified live against the 802.1 (ICne) document store on 25 July 2026.
- Bilateral Transactions in Ethernet: Eliminating Timeout-and-Retry at the Source — Paul Borrill. DCN 1-26-0015-00-ICne (Tuesday 14 July). Mentor DCN 0015. Predecessor: Availability Is a Bilateral Property, DCN 1-26-0012-02.
- Defining Atomicity for the Link Layer: From Unilateral Send to Bilateral Commit — Paul Borrill. Paper DCN 1-26-0039-00-ICne, slides DCN 1-26-0041-00-ICne (Thursday 16 July). The proposed new definition of atomicity for IEEE 802.1 and the conceptual anchor for the proposed Nendica Study Item Atomic Link Semantics for AI Computing Networks (ALS-AICN). Mentor DCN 0039 · slides DCN 0041.
Network Efficiency from First Principles — Parts 1–12. The twelve-part series derives, quantifies, and validates the pathologies the two headline contributions propose to remove. Indexed one-page as the “At a Glance” summary, DCN 1-26-0033-00-ICne.
| DCN | Title | Link |
|---|---|---|
| 0016-01 | Part 1 — The Address-Redundancy Tax | Mentor → |
| 0017-01 | Part 2 — The Fixed-Framing and Coding Tax | Mentor → |
| 0018-01 | Part 3 — UALink, A Shipping Case Study | Mentor → |
| 0019-01 | Part 4 — The Other Fixed Bytes: Preamble, Delimiter, Gap | Mentor → |
| 0020-01 | Part 5 — Right-Sizing the Coding Tax with PIF | Mentor → |
| 0021-01 | Part 6 — Timeout-and-Retry Storms | Mentor → |
| 0024-00 | Part 7 — The Cost of Confirming | Mentor → |
| 0025-00 | Part 8 — Lossless, Deadlock-Free, Any Topology: Pick Two | Mentor → |
| 0026-00 | Part 9 — The Split-Brain Window | Mentor → |
| 0027-00 | Part 10 — Issues at Hyperscale: A Response to Hoefler | Mentor → |
| 0028-00 | Part 11 — The Shannon Ceiling | Mentor → |
| 0029-00 | Part 12 — Harvesting Instead of Dissipating: Recoverable Link Energy | Mentor → |
Netdev 0x1A — Bilateral Commit at Two Layers (Rome, 13–16 July 2026)
Two peer-reviewed papers were presented on the Moonshot track at Netdev 0x1A (THE Technical Conference on Linux Networking), Università Roma TRE. Both are public (no login). Netdev mints no DOI; cite the session page.
- Why Syscalls Fail: Bilateral Commit at the Linux Kernel/Userspace Boundary — Manav Singhai and Paul Borrill (Moonshot #4). Introduces the Kernel Acknowledgment Spectrum (L0–L4) and shows that most syscall interfaces terminate at L1 or below, so failures (OOM kills,
fsync,io_uringcompletion races, memory overcommit) are silent. session page · final PDF · slides - Promise Networks: Why a Bilateral Link Layer Solves Congestion Control at the Source — Anjali Singhai-Jain, Paul Borrill, Chihjen Chang, and David Zage (Moonshot #46). Generalises Metcalfe & Boggs’s 1976 EFTP “end-dally” into a link-layer admission primitive in which every frame is gated on a peer-issued token, so uninvited frames never enter the fabric. session page · final PDF · slides
Session recordings are not yet posted — Netdev uploads them to its YouTube channel after the event; each session page will carry its video when it goes live.
Important Links
Back to Open Atomic Ethernet homepage
OAE Zoom Room
https://opencompute-org.zoom.us/j/89756960348?pwd=vKwSvuFHzSCQsVEpHI36fAJQCeXez8.1
Github
Charter
- Æthernet: Open-SmartNIC. Non-proprietary, formally verifiable API, stack and hardware implementation.
- Chiplet Interconnects
- In-rack communication
Workstreams
| Project | Objective | Lead | Meeting Schedule | Status | |
|---|---|---|---|---|---|
| Plenary meetings are suspended until after Chiplet Summit 2026 | |||||
| #1 | Physical, Logical, and Virtual Configuration “PLV” | Establishing secure enrollment, key exchange, and verifiable network topology for OAE. Open discussions on Physical Configuration [boot-up security], Logical Configuration [tenant confinement, Kubernetes], Virtual Configuration for Distributed Applications built on the Open Atomic Ethernet protocol. Graph Theory Confinement Models built on DAGs, Trees, and Graph Covers. | Paul Borrill (DÆDÆLUS) | Monday 8am PT weekly | Work in progress |
| #2 | Cognitive Networking “COGN” | Open source Implementations of the Open Atomic Ethernet protocol. | Kevin Cameron | Inactive | Work in progress |
Whitepapers
OPEN ATOMIC ETHERNET, THE TWO GENERALS PROBLEM, AND FABRIC ATOMICITY GAPS
WHY LINK FLAPPING IS A SEMANTIC PROBLEM, NOT A HARDWARE PROBLEM
Open Atomic Ethernet SerDes Specification
Atomic Interconnects Without Timeouts: A Comparative Technical Analysis
Open Atomic Ethernet Comprehensive Engineering Analysis (Jan 2026)
State Machine Proposal (3/5/25)
Reinventing Reliability at Layer 2
Foundational Documents
- Ethernet-Boggs-Metcalfe.pdf
- ReliableTransmissionOverHalfDuplex_Lynch_CACM_1968.pdf
- NoteOnReliableTransmissionOverHalfDuplex.pdf
Weekly Plenary Meeting
- Wednesday 9am – 10am PT
Zoom Details
OCP TAP – Open Atomic Ethernet (OAE)
- Meeting ID: 897 5696 0348
- Passcode: 656480
One tap mobile
- +16469313860,,89756960348# US
- +19292056099,,89756960348# US (New York)
Dial by your location
- +1 929 205 6099 US (New York)
- +1 301 715 8592 US (Washington DC)
- +1 312 626 6799 US (Chicago)
- +1 253 215 8782 US (Tacoma)
- +1 346 248 7799 US (Houston)
- +1 669 900 6833 US (San Jose)
- Find your local number: https://opencompute-org.zoom.us/u/kbjKzDcBqY
Upcoming Calls
Legend
- OAE = Open Atomic Æthernet
- OCP = Open Compute Project
- TAP = Time Appliances Project
| Call | Date | Topics | Speakers |
|---|---|---|---|
| TAP #151 | Jan-28, 2026 | Time is not a conserved quantity | Paul Borrill, Ph.D. |
Meeting History
| Index | Date | Topics | Speaker | Slides+Artifacts | Summary |
|---|---|---|---|---|---|
| #113 PLV #53 | Mar-02, 2026 | Cloud-Native & Kubernetes Replacement [More] | Victor Lu | CNCF Slides Chat Transcript | Victor Lu reviewed CNCF Cloud-Native concepts. Paul Borrill detailed how OAE’s Graph Virtual Machine replaces Kubernetes infrastructure for secure deployments. A debate on business models led Paul to cancel future meetings to focus on technical innovation.[Full Summary] |
| #112 PLV #52 | Feb-23, 2026 | Circumventing FLP & iCloud Synchronization Failures [More] | Paul Borrill | Summary Transcript Chat Why iCloud Fails FLP & OAE Graph Virtual Machine | Paul Borrill analyzed iCloud synchronization failures and proposed circumventing the FLP impossibility result using OAE’s three-party failover system and epistemic registers. The team discussed replacing Kubernetes with the Graph Virtual Machine (GVM) in secure cellular fabrics.[Full Summary] |
| #111 OAE #50 | Feb-09, 2026 | Architecture Trumps Algorithms: Why Faster Shortest-Path Calculation Won’t Save Your Network [More] | Paul Borrill & Dean Gladish | Summary Transcript Technical Report Mathematica .nb Chat Dijkstra-FITO-TR | Paul Borrill identified the focus on SPF speedups as a category error, noting that epistemic uncertainty in FITO protocols—not computation—limits convergence. Dean Gladish demonstrated self-healing communication trees and SerDes simulations for scale-invariant, timeout-free reliability.[Full Summary] |
| #110 OAE #51 | Jan-21, 2026 | Phantom Nodes & Link Flapping [More] | Paul Borrill | Summary Transcript Slides | Paul Borrill announced organizational changes and a break until Feb 25 for the Chiplet Summit. The technical inquiry addressed “Phantom Nodes” and “Link Flapping” in AI clusters, arguing that these are semantic failures of Forward-In-Time-Only (FITO) protocols. OAE solves this via Perfect Information Feedback and Triangle Networks, eliminating timeouts (TAR) for deterministic reliability. [Full Summary] |
| #109 TAP #149 | Dec-31, 2025 | From Timeouts to Transactions for Deterministic, Observable Layer-2 Networking [More] | Paul Borrill, Ph.D. | Slides | OAE 2025 progress: replace TAR with transactional L2 using reversible echo + triangle supervision for deterministic, timeout-free link reliability; zero-admin secure bring-up; graph-based scouting/topology for chiplet & datacenter fabrics.[Full Summary] |
| #108 CAS #41 | Dec-08, 2025 | Chiplet Fusion & Firmware-Free Boot [More] | Paul Borrill | Summary Transcript Chat Chiplet Summit | CAS #40 prioritized Chiplet Summit strategy, targeting IEEE 802.3 and Ethernet Alliance leaders for spec review. Paul Borrill defined OAE’s “Firmware-Free Boot”—an Oxide Computer-inspired model that bypasses UEFI by booting chiplets directly into a RISC-V microkernel, subsuming legacy configuration layers. The group resolved to move meetings to 9 AM in the new year. [Full Summary] |
| #107 CAS #40 | Dec-01, 2025 | Eulerian Scouting & Topology Discovery [More] | Paul Borrill | Summary Transcript Chat Eulerian Path | CAS #40 welcomed Eric Rentschler (Molex) and Russ Wunderlich (OCP). The session defined the Topology Discovery phase, choosing Eulerian path traversal (visiting every link) over Hamiltonian paths to guarantee complete network mapping in polynomial time. Paul Borrill described an “ant/scout” mechanism using breadcrumbs/pebbles (port-local state) to execute loop-free Tremaux exploration, while David Johnston enforced that discovery must strictly follow neighbor authentication to establish a trusted domain. Dugan Hammock’s simulations favored “snakes” (continuous streams) over single ants for faster convergence. [Full Summary] |
| #106 CAS #39 | Nov-24, 2025 | IBE-PUF & Bring-Up [More] | David Johnston | Summary Transcript Chat Efficient Key Exchange Using Identity-Based Encryption | CAS #39 defined the Network Bring-Up Sequence—ordering Link Liveness → Topology Discovery → Distributed Firing Squad → IBE Setup—and endorsed PUF+KDF architectures for stable, identity-based encryption. John Lockwood urged the group to layer OAE’s transactional integrity atop UA Link to leverage industrial momentum, shifting the strategy from competing on the wire to unifying the logical interconnect. [Full Summary] |
| #105 CAS #38 | Nov-17, 2025 | Lightweight Crypto & Chiplet Boot [More] | David Levy | Summary Transcript Chat Network Security: A Systems Approach | CAS #38 introduced David Levy to discuss ASCON as a lightweight cryptography standard for OAE’s constrained chiplet environment. David Johnston presented a low-resource RNG architecture—chaining a BIW extractor ($A times B + C$) into an ASCON XOF—and detailed Identity-Based Encryption (IBE) rooted in HCI-based PUFs to enable pairwise key derivation without negotiation. Paul Borrill outlined the System Management lifecycle, spanning Firing Squad-based boot sequences to runtime Triangle Network failover, establishing OAE as a superior, open alternative to NVLink. [Full Summary] |
| #104 CAS #37 | Nov-10, 2025 | Tang Nano Link Engine [More] | David Johnston | Summary Transcript Chat | CAS #37 dissects OAE’s FPGA link engine as a concrete testbed: David Johnston’s SciSpeed Tang Nano 20K design now streams 64-bit words over a quadrature request/acknowledge protocol with source-synchronous clocking and a maximal-length 32-bit LFSR/CRC core. The team used this to motivate a state-vector and matrix view of alternating causality—tokens as indivisible state—and explored Verilog→C++ translation for symbolic/formal verification. Operationally, they decided to cancel overlapping OAE plenaries, shift to monthly plenary meetings, and use the link engine plus Mathematica state-vector models as the canonical reference for CAS.[Full Summary] |
| #103 CGN #27 | Nov-06, 2025 | Rust Chip Spec [More] | Paul Borrill | Summary Transcript | CGN #27 formalized the team around a minimal Rust chip specification as the seed of an executable OAE stack, while David Johnston focused on completing the link engine and key-distribution text. The group scoped a Mathematica ping-pong simulator with hazard injection (Dugan + Dean) and formalized Sahas Munamala’s work on the firing squad and transition byte as inputs to the main spec, alongside a recruiting push for RTL/FPGA experts and potential partners such as Algo-Logic. [Full Summary] |
| #102 OAE #40 | Nov-05, 2025 | Photon Mirrors to FPGA [More] | Sahas Munamala | Summary Transcript Chat | OAE #40 combined Sahas Munamala’s alternating causality photon-mirror model with concrete Bits and Bytes and minimal routing work, showing how reversible forward/reverse operators lead to information-conserving channels and an executable, NVLink-style 64-byte framing of the OAE protocol. The group converged on identity + randomness–based secure neighbor bring-up, a shift to monthly plenaries with weekly CAS workstreams, a holiday push to convert provisional patents to full filings, and assigned Dean Gladish responsibility for finishing the OAE wiki as the public face of spec v7A. [Full Summary] |
| #101 CAS #36 | Nov-03, 2025 | Slice-Based OAE Spec [More] | Sahas Munamala | Sahas FPGA Substack “Unnecessary Problems: X-Propagation” Summary Transcript Chat | CAS #36 advanced OAE’s link protocol from informal discussion toward an executable specification. Sahas Munamala’s UART-on-GOWIN work and David Johnston’s FPGA experiments framed the need for a functional stack where message processing is described once in a human- and machine-readable form. David pivoted away from hierarchical IBE toward elliptic-curve pairings and a compact CERT profile intended to pass OpenSSL, while he and Kevin Cameron debated P4, SystemVerilog, and ITU-style state machines for defining both C-level interfaces and on-the-wire bit patterns. The meeting concluded with a GOWIN IDE demo and concrete next steps: a low-cost link demonstrator, slides of the physical setup, and a cleaned-up cert specification. [Full Summary] |
| #100 CGN #26 | Oct-30, 2025 | 1,100 RISC-V Ops to TSVF [More] | Paul Borrill | Hebbian Learning and the LMS Algorithm | Paul shared RISC-V insights—1100+ instructions and 144 extensions—and described mapping FFT algorithms onto the GVM/Octovalent stack. He introduced the two-state vector formalism as a physics-based backpropagation using forward/ backward time for perfect information transfer, surpassing CRC/FEC methods. David Johnston noted OAE’s cryptographic bring-up collapses Azure’s first three phases—identity, addressing, and discovery—while Kevin Cameron cited Bernard Widrow’s work on neural fault tolerance, suggesting AI-driven resilience for OAE. [Full Summary] |
| #99 OAE #39 | Oct-29, 2025 | Unified L2 Stack [More] | Paul Borrill | Summary Transcript Chat | Paul proposed a joint OCP–RISC-V–Chiplet Summit initiative to tackle the chiplet yield problem, drawing on lessons from Sun Microsystems’ bring-up challenges. Pankaj Mehra cautioned the scope blurred data center and EDA domains, prompting Paul to clarify it as a Layer-2, first-principles redesign based on David Johnston’s cryptographic bring-up and scouting protocols. David noted the approach could collapse Azure’s 11-phase bring-up to one, and Paul urged OCP to adopt LaTeX for specifications following the RISC-V documentation model. [Full Summary] |
| #98 CAS #35 | Oct-27, 2025 | GoWin Mesh [More] | Sahas Munamala | Summary Transcript | CAS #35 refined OAE’s minimal hardware design, confirming liveness and two-phase commit as a single intertwined state machine with a defined stall state. The bring-up was formalized as an observer-free process using jitter settling and monotonic counters instead of timeouts. David Johnston presented Hierarchical IBE for layered keying and showed progress on a low-cost GoWin/RISC-V FPGA mesh demonstrator. [Full Summary] |
| #97 CGN #25 | Oct-23, 2025 | $35 RISC-V FPGA [More] | Sahas Munamala | P4 Workflow $35 FPGA Board Summary Transcript Chat | CGN #25 set a milestone for a low-cost RISC-V FPGA demo implementing OAE’s link-layer core. David Johnston identified a $35 board for liveness and two-phase commit, avoiding costly hardware. Sahas Munamala will draft the one-page spec, while Kevin Cameron emphasized clean SmartNIC–software integration to align with OAE’s transactional model. [Full Summary] |
| #96 CAS #34 | Oct-20, 2025 | “Christmas Trees” to Dual SAPs [More] | Paul Borrill | Stanford Paper Summary Transcript Chat | CAS #34 focused on the OAE SmartNIC and its architecture. The team endorsed RISC-V for its open, register-to-register design and adopted Paul’s modular “skyscraper” API concept, with David Johnston stressing clear service boundaries like 802 MSAPs. Debate over legacy support ended with a hybrid model—retaining a standard Ethernet MSAP while adding a new transactional SAP for OAE. [Full Summary] |
| #95 CGN #24 | Oct-16, 2025 | Laser Gyros & 4-Phase Commit [More] | Paul Borrill | Summary | Sahas presented his *“photon mirror”* paper, introducing the PIF model and a framework for alternating causality—drawing parallels to laser gyroscopes. Paul used this to challenge networking’s reliance on forward-only time and end-to-end principles, advocating for a *four-phase commit* grounded in physical reversibility. The group extended the discussion to entropy limits, hardware programmability, and DÆDÆLUS’s *IP strategy*: open software infrastructure protected by patents on its link-level mechanisms. [Full Summary] |
| #94 CGN #23 | Oct-09, 2025 | Spinnaker Project: 2D Neuromorphic Surface [More] | Kevin Cameron | Summary | The Cognitive Networks team explored integrating _Spinnaker_’s 2D compute surface with _Open Atomic Ethernet (OAE)_ to demonstrate containerized workloads on a fully open communication fabric. Kevin Cameron proposed using Spinnaker chiplets as a container-native substrate, while Paul framed OAE as a firmware-free, model-driven boot process. The collaboration aims to showcase OAE-enabled container communication on _Spinnaker 2_, laying groundwork for an open RISC-V–based _Spinnaker 3_ stack from API to wire. [Full Summary] |
| #93 OAE #36 | Oct-08, 2025 | ACL / IAM Failures to Capability Certificates [More] | Alan Karp | Summary | Alan Karp critiqued ACLs and authentication-based IAM systems for delegation flaws like the confused deputy problem and weak revocation. He proposed a capability-based model using signed certificates as transferable tokens of authority. The group discussed integrating this via IBE keys and Layer-2 tree/DAG structures to constrain delegation while preserving reversibility and control. [Full Summary] |
| #92 CAS #32 | Oct-06, 2025 | OAE Spec: GVM, Surprisal Error Model[More] | Paul Borrill | Summary | CAS #32 focused on OAE’s security and validation. David Johnston proposed compact ED25519 certificates (<256 bytes) to replace bulky X.509s and will draft an OAE-specific profile. The team validated Sahas Munamala’s spec capturing reversible state machines, Link Lambdas, the GVM, and surprisal-based fault detection. They also addressed GPU/NVLink reliability issues, proposing OAE’s atomic firing squad protocol for atomic, continuous snapshots and updates. [Full Summary] |
| #91 CGN #22 | Oct-02, 2025 | 8-Bit Liveness Byte [More] | Paul Borrill | Summary Transcript | OAE defined Link Lambdas as a finite hardware set of reversible primitives and the Graph Virtual Machine (GVM) as their programmable orchestrator. The group standardized terminology—renaming the Protocol Byte to Operator Byte—and confirmed the 8-bit Liveness Byte as essential to the Truncated Tail Latency protocol, ensuring bidirectional awareness and removing the need for timeouts or retries. [Full Summary] |
| #90 OAE #35 | Oct-01, 2025 | Atomic Swaps, No Shared Memory [More] | Paul Borrill | Summary Transcript Chat | The session reaffirmed OAE’s reversible, register-to-register design, where all operations are atomic swaps enabling lossless rollback without timeouts. It also proposed adapting SCTP’s four-way handshake for Layer 2 to provide authenticated, address-free connectivity with inherent DDoS resistance. [Full Summary] |
| #89 CAS #31 | Sep-29, 2025 | Two-Layer Hybrid IBE–PKI [More] | David Johnston | Unicron Paper | CAS established a dual-plane keying model: Identity-Based Encryption (IBE) secures tenant data for confidentiality even from operators, while Public Key Infrastructure (PKI) governs management actions under a root of trust—balancing cryptographic rigor with interoperability and deployment pragmatism. [Full Summary] |
| #88 CGN #21 | Sep-25, 2025 | Reversible Link Lambdas [More] | Sahas Munamala | Summary Transcript Chat | Link Lambdas were formalized as reversible in-network compute functions; Computation Tree Logic (CTL) was adopted as the GVM’s temporal-spatial logic; and Reliable Address-Free Ethernet (RAF) was defined to confine IP addressing to the fabric’s holographic boundary, securing interior nodes while preserving legacy interoperability. [Full Summary] |
| #87 OAE #34 | Sep-24, 2025 | Edward A. Lee; Plato & The Nerd [More] | Edward A. Lee | Summary Transcript Chat | Edward A. Lee’s plenary reinforced OAE’s philosophical foundation: engineering reshapes reality to fit abstract models, not vice versa. He challenged pan-computationalism’s determinism and affirmed that standards can be executable and formal, supporting OAE’s first-principles goal of mathematically verifiable specifications grounded in reversibility. [Full Summary] |
| #86 CAS #30 | Sep-22, 2025 | YANG over NetConf to Redfish [More] | David Johnston | Summary Transcript Chat | CAS adopted YANG-based modeling to define OAE’s configuration hierarchy and formalized transactional semantics via reusable profiles, decoupled from legacy standards. The group also outlined a self-organizing, immune-style network—combining fisheye topology awareness for scalable coordination and consensus-driven defense against compromised nodes. [Full Summary] |
| #85 CGN #20 | Sep-18, 2025 | 31-Bit Operator to NVIDIA [More] | Paul Borrill | Summary Transcript Chat | Deep dive formalized a reversible 31-bit operator for link transactions, partitioned as `2+8+8+1+12=31` bits → protocol type, link liveness, state-machine ID, an ancilla (causal orientation) bit, and a state-transition field; endorsed slice-level ACKs to exploit sub-microsecond propagation for fine-grain transactional control; examined three-state (forward/backward/error) encoding via the ancilla mechanism; closed with NVIDIA outreach planning and an upcoming Link Lambdas executable-spec demo. [Full Summary] |
| #84 CAS #29 | Sep-15, 2025 | Link Lambdas [More] | David Johnston | Summary Transcript Chat | Spec emphasized token-carried Link Lambda state—including a forward/backward ancilla bit—so the link behaves as one unified state machine; security adopts optional confidentiality with mandatory authentication/integrity (GMAC), requiring a full key exchange even for integrity-only modes; constant-time ECC P-384 was advanced to mitigate timing side-channels. [Full Summary] |
| #83 CGN #19 | Sep-11, 2025 | Ciphers to Chip [More] | Sahas Munamala | Requirements Spreadsheet Summary Transcript Chat | Spec v0.7 progressed from concepts toward verifiable hardware: refined abstraction, scoped Link Lambdas, and debated whether reversibility is mandatory; it also called for a formal, machine-readable language and examined local discovery for unknown orientation alongside protocol trade-offs. [Full Summary] |
| #82 OAE #30 | Sep-10, 2025 | v0.7 Plenary [More] Transcript Chat | Sahas Munamala | Slides OAE-SPEC-MAIN Summary | Plenary v0.7 update: core chapters were formalized; two new pillars—Link Lambdas (in-NIC, application-defined functions) and Compute Tree Logic (capability-scoped provisioning)—were introduced; the group pushed for a formally verified Layer-2 management protocol; and clarified a unified, packet-carried link state that enables deterministic rollback. [Full Summary] |
| #81 CAS #11 | Sep-8, 2025 | Video Transcript Chat | David Johnston | Zoom Recap | MIBs and YANG: Management Configuration. 802 management review: migration from SNMP to NETCONF/YANG; evaluation of configuration models and data formats; exploration of potential new Layer-2 standards; closed with space-mission datacenter needs and Atomic Ethernet protocols emphasizing security and efficiency. [Full Summary] |
| #80 CGN #18 | Sep-4, 2025 | Cross-Fabric Dialogue [More] | Sahas Munamala | Audio Summary | Strategic outreach: Sahas invited PCIe/CXL expert Annie Liao to engage with OAE as a fixed-radix fabric, aiming to bring fresh outside perspectives and deepen dialogue across competing interconnect communities. [Full Summary] |
| #79 OAE #29 | Sep-3, 2025 | N/A (Canceled) | [More] | N/A (Canceled) | Canceled for reorganization and speaker scheduling. [Full Summary] |
| #78 CGN #17 | Aug-28, 2025 | Reconfigurable Systems Transcript Chat | Kevin Cameron | Zoom Recap | Dynamic Recompilation & Reconfigurability. Meeting covered programming models/specs and OSS–proprietary integration; explored network-aware OS directions with dynamic recompilation, power optimization, and SmartNIC offload; and addressed security models, hardware failover, and automated orchestration, closing with collaboration plans and next-step agendas. [Full Summary] |
| #77 OAE #28 | Aug-27, 2025 | David Johnston on Entropy [More] | David Johnston | Slides Summary Transcript Chat | Guest talk: David Johnston surveyed entropy beyond Shannon—introducing Rényi measures (min-entropy for cryptography, Hill entropy for bounded adversaries)—highlighted pitfalls from serially correlated noise, explained quantum-secure two-source extractors behind NIST RNGs, and linked information erasure to energy via Landauer’s principle. [Full Summary] |
| #76 CAS #26 | Aug-25, 2025 | RLNC-Inspired OAE [More] | Paul Borrill | Network Coding Meets TCP Reversible Transactions Demo (2023) Summary Closed Captions Transcript Chat | CAS evaluated Medard-style network coding for OAE, showed a reversible DÆDÆLUS FPGA demo with ≈220 ns round-trip and deterministic rollback, and debated “always-on” reversible links versus sleepy-link energy models under Landauer’s principle. [Full Summary] |
| #75 CGN #16 | Aug-21, 2025 | From Apps to Silicon [More] | Kevin Cameron | LoongArch ISA Lambda Calculus in 383 Bytes Asynchronous Game of Life Transcript Chat | Software-driven silicon flow: decompose workloads into message-passing components, replace shared memory with asynchronous messaging, and use executable specifications to compile behavior into verifiable FPGA/chiplet hardware. [Full Summary] |
| #74 OAE #34 | Aug-20, 2025 | Pappas on Standards [More] | Jim Pappas | Summary Closed Captions Slide Transcript Chat | Jim Pappas’s standards “playbook”: win by extreme focus—avoid “Christmas-tree” feature bloat, unblock the one issue that stalls adoption, and drive costs relentlessly down; exemplified by USB’s dual-speed compromise for EMI and saving cents on connectors to unlock scale. [Full Summary] |
| #73 CAS #25 | Aug-18, 2025 | Surprisal, Not Checksums [More] | Paul Borrill | Meeting Assets Transcript Chat | OAE error handling replaces conventional EDC/FEC with reversibility and surprisal-based detection in a tightly coupled link-state machine, targeting a formally verifiable specification that enumerates all recovery states, including coordinator-led, deterministic roll-back. [Full Summary] |
| #72 CGN #15 | Aug-14, 2025 | Software-Driven Silicon [More] | Paul Borrill | Photonics Efficient Computer Asynchronous Game of Life Synchronous Game of Life Transcript Chat | Bridged high-level software design to verifiable hardware via a software-driven silicon flow: decomposed workloads into compute/communication structures, replaced shared memory with asynchronous message passing, and used executable specs (e.g., an async Game of Life) to compile behavior into optimized FPGA/chiplet designs. [Full Summary] |
| #71 OAE #33 | Aug-13, 2025 | FMS: Reversible CAS [More] | Sahas Munamala | Meeting Assets Slides Transcript | FMS talk: OAE makes the network safe for transactions via reversible compare-and-swap on interlocked link-state machines and a fixed 64-byte frame (8×8-byte slices), contrasting Clos with a resilient direct-connect mesh; discussion covered TSN incompatibility, fixed-size mapping, and head-of-line mitigation. [Full Summary] |
| #70 CAS #24 | Aug-11, 2025 | CAS Spec Foundations [More] | David Johnston | MITRE ATT&CK Framework SummaryFinite Automata Document Graph Report Transcript Chat | CAS codified spec foundations: a JAM bit for packet typing, in-packet ancilla state to unify the link state machine, PROMELA-style formal modeling for verifiability, and a MITRE ATT&CK–based threat model. [Full Summary] |
| #69 OAE #32 | Aug-06, 2025 | N/A (Canceled) [More] | N/A (Canceled) | Mathematica-verified simulations show OAE’s direct-connect mesh outperforms Clos for east–west traffic, degrades gracefully under faults, and delivers Layer-2 exactly-once via reversible, slice-level ACK transactions. [Full Summary] | |
| #68 CAS #23 | Aug-04, 2025 | CAS [More] | David Johnston | Summary Transcript Chat | Machine-readable configuration language with FSP-based synchronous updates; reversible _swap_ transactions supersede get/set; multi-node “sync groups” tied to consensus (e.g., Paxos/FSP); timing grounded in logical simultaneity and reversibility; continued graph-FSP progress and a multi-packet GCM link cipher. [Full Summary] |
| #67 CGN #14 | Jul-31, 2025 | Link Cipher & Formal Verification [More] | David Johnston | Summary Transcript Chat | OAE’s optional cipher uses multi-packet AES-GCM (MPGCM) with SHF/STF framing and a monotonic IV to ensure nonce safety under reversal. Key rollovers are modeled as the Firing Squad Problem and verified in Spin via PROMELA. A unified graph language and GVM enable formal deployment of ML/AI topologies like HammingMesh and virtual toroids from a single Mathematica model. [Full Summary] |
| #66 OAE #31 | Jul-30, 2025 | FMS Presentation Practice [More] | Sahas Munamala | .cc.vtt Summary Transcript Chat | FMS dry run contrasted ACID with today’s unreliable networks and positioned OAE as a Layer-2 transactional fabric delivering exactly-once semantics. Demos showed a self-healing direct-connect mesh and a 50-node emulator that reroutes on link failure. Feedback emphasized anchoring the talk on a single, decisive concept: you can’t add reliability atop unreliability. [Full Summary] |
| #65 CAS #21 | Jul-28, 2025 | Requirements & Cost of Reliability [More] | Paul Borrill | Summary Transcript Chat | Requirements debate crystallized audience and problem framing: reduce the cost of reliability by delivering a transactional Layer-2 mesh of XPUs (“determinism on demand”), with an action to publish a canonical, community-visible requirements document. [Full Summary] |
| #64 CGN #13 | Jul-24, 2025 | CGNitive Networks Discussion [More] | Paul Borrill | Summary Time-Relativity-Ethernet.pdf Transcript Chat | Physics Discussion [Full Summary] |
| #63 OAE #30 | Jul-23, 2025 | Explore Mathematica Sim. Tool v2 [More] | Dean Gladish | Summary Transcript Chat | The plenary showcased Mathematica as code-as-proof, with executable notebooks modeling reversible dynamics and a Clos-vs-Dædælus topology comparison. It reframed failures as artifacts of a single global timeline and concluded with a live bipartite-link “circulating snakes” demo using 8-byte slice tokens and elastic bandwidth. [Full Summary] |
| #62 CAS #21 | Jul-21, 2025 | Configuration Language and Swaps [More] | David Johnston | Summary Transcript Chat | Debated a canonical, machine-readable config (ASN.1 vs JSON); endorsed atomic “swap” transactions for reversible integrity; generalized to multi-node sync groups via graph-based Firing Squad consensus; rejected wall-clock dependence in favor of causal, reversible protocols that avoid timeout-driven failure modes. [Full Summary] |
| #61 CGN #18 | Jul-17, 2025 | Wakesman Firing Squad Problem [More] | David Johnston | Summary Transcript Chat | Explored transformers-as-compilers and side-channel risks, emphasizing formal equivalence; proposed a reliable, reversible protocol to supplant timeout/retry recovery and Raft-backed stores like etcd in distributed systems. [Full Summary] |
| #60 OAE #29 | Jul-16, 2025 | Explore of Mathematica as a Spec. Tool [More] | Sahas Munamala | Slides Summary Transcript Chat | Clarified network (hop-by-hop) vs application (end-to-end) responsibilities; defined reversibility as deterministically running the state machine backwards; agreed on a hybrid simulation plan—Mathematica for protocol experiments and NS-3 for packet-level hazards. [Full Summary] |
| #59 CAS #20 | Jul-14, 2025 | PROMELA Modeling–DJ [More] | David Johnston | Slides Summary Transcript Chat | CAS #20 established core OAE configuration requirements; rejected 802-style MIBs and conventional IaC as ill-suited to container ergonomics; assigned David Johnston to draft a rapid-discovery specification. [Full Summary] |
| #58 CGN #17 | Jul-10, 2025 | HammingMesh [More] | Paul Borrill | Paper Zoom Recap Transcript ChatDraft 0.3 Live Google Doc | Shift from FPGA work to self-aware systems targeting deep-learning network topology; review of HammingMesh and requirements; debate on TCP alternatives, VLAN design, and containerized deployment strategy. [Full Summary] |
| #57 OAE #28 | Jul-09, 2025 | Corpus CLI [More] | Sumanth Pallamreddy | Slides Zoom Recap Transcript Chat | Corpus CLI introduces an AI-driven specification-analysis workflow: RAG-based rapid indexing, semantic querying, and automated diagram generation that accelerate research and comprehension of dense standards (demonstrated on OAE). [Full Summary] |
| #56 CAS #19 | Jul-07, 2025 | Refining the Configuration Model [More] | David Johnston | Zoom Recap Transcript Chat | Configuration vs. discovery was clarified; dedicated high-priority configuration packets were proposed; and the protocol must support both Firing-Squad-style synchronous network-wide updates and node-local asynchronous changes, with an action to fold feedback into the configuration charter. [Full Summary] |
| #55 CGN #16 | Jul-03, 2025 | Virtual Channels & App-Aware Network [More] | Paul Borrill | Summary Transcript Chat | Debate on link-layer atomicity: Kevin placed atomicity at the application layer; Paul countered that exactly-once semantics requires conservation of information and per-link Alternating Bit Protocol (ABP) to preserve the spacetime handshake and avoid epistemic loss. [Full Summary] |
| #54 OAE #27 | Jul-02, 2025 | A Random Walk Through the Ether [More] | Bijan Nowroozi | Slides Zoom Recap Transcript Chat | Keynote and progress update: ~18-day slip and secure-enclave roadmap; RDMA/Derecho, MPI-SHARP, and SCTP mapped onto OAE’s slice-based MAC, indicating ~10× lower micro-burst latency and clear paths to hop-aware congestion control, atomic multicast, and reversible transactions (toward a deterministic, exactly-once fabric). [Full Summary] |
| #53 SAR #18 | Jun-30, 2025 | Physically Unclonable Functions [More] | David Johnston | Zoom Recap Transcript Slides | Silicon-rooted identity via PUFs underpins secure links: engineer stable, reproducible device responses from manufacturing variation; mitigate entropy loss and noise during enrollment; and design hardware-bound authentication protocols—advancing an intrinsic, verifiable root of trust for secure, reversible transactions. [Full Summary] |
| #52 CGN #15 | Jun-26, 2025 | Summary [More] | Paul Borrill | Formal methods were endorsed to achieve “determinism on demand”: adopt Promela/Spin to model causality across multiple time contexts, define a formal specification language to prove implementation guarantees, and pursue follow-ups to document industrial and formal-verification challenges. [Full Summary] | |
| #51 OAE #20 | Jun-25, 2025 | Making a New Ethernet [More] | David Johnston | Zoom Recap Introduction Slides David’s Final Set Transcript Chat | David Johnston outlined the practical path to a new IEEE 802 Ethernet standard—highlighting physical-layer tradeoffs (optical vs. electrical), power budgets, protocol timing, and link-layer security requirements. [Full Summary] |
| #50 Rehearsal | Jun-24, 2025 | Summary [More] | Paul Borrill | A LinkedIn formatting error introduced a line break that broke the OAE #20 Zoom URL. Editing the post to render the URL as a single, continuous string restored access. OAE Zoom meetings remain open to the public. [Full Summary] | |
| #49 TAR #17 | Jun-23, 2025 | Topology & Routing [More] | Paul Borrill | Zoom Recap Transcript Chat | Version 0.6 defines a first-principles, non-incremental architecture that prioritizes conceptual clarity over legacy compatibility. It replaces Kubernetes with a Graph Virtual Machine, introduces biologically inspired routing and the P8 interaction language, and optimizes mobile and containerized workloads end-to-end. [Full Summary] |
| #48 CGN #14 | Jun-19, 2025 | Discussion [More] | Paul Borrill | Zoom Recap Transcript Chat | Chiplet Ethernet [Full Summary] |
| #47 OAE #19 | Jun-18, 2025 | Plenary Discussion [More] | Sahas Munamala | Zoom Recap Transcript Chat | Spec Progress [Full Summary] |
| #46 TAR #16 | Jun-16, 2025 | Topology & Routing [More] | Sahas Munamala | Zoom Recap Packet Spraying Transcript Chat | UEC: Packet Spraying [Full Summary] |
| #45 CGN #13 | Jun-12, 2025 | Discussion [More] | Paul Borrill | Zoom Recap Transcript Chat | Kevin’s protocol proposal and clarifying a language for describing communication requirements that can map across diverse chiplet-based hardware. The group examined challenges in chiplet architectures—manufacturing defects, routing, buffer management, and virtual addressing—while arguing for a non-hierarchical, graph-based network that can discover topology, survive faults, and deliver reliability beyond “best-effort” Ethernet. They closed by identifying Kevin’s Samsung deployment as a promising test bed for more reliable, transaction-oriented distributed systems and by outlining follow-up work on protocol stack definition, documentation, and shared materials. [Full Summary] |
| #44 OAE #18 | Jun-11, 2025 | Plenary Discussion [More] | Andy Helland | Zoom Recap Transcript | Market Discussion [Full Summary] |
| #43 TAR #10 | Jun-9, 2025 | Topology & Routing [More] | Paul Borrill | Zoom Recap | Machine Learning Operations Framework [Full Summary] |
| #42 TAR #9 | Jun-2, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Transcript Chat | Flow Control & Infiniband [Full Summary] |
| #41 TAR #12 | May-19, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Transcript Chat | P8 Programming [Full Summary] |
| #40 OAE #15 | May-14, 2025 | Plenary Discussion [More] | Paul Borrill | Zoom Recap Transcript | Market Discussion [Full Summary] |
| #39 TAR #11 | May-12, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Transcript Chat | P8 Programming [Full Summary] |
| #38 CGN #8 | May-08, 2025 | Discussion [More] | Paul Borrill | xUI Paper Transcript Chat | xUI — User Interrupts [Full Summary] |
| #37 OAE #14 | May-07, 2025 | Plenary Discussion [More] | Paul Borrill | Zoom Recap Transcript Principles of Operation v0.1 Æthernet and Switched Networks v0.3 | Document Review [Full Summary] |
| #36 TAR #10 | May-05, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Transcript | HPC Environment [Full Summary] |
| #35 CGN #7 | May-01, 2025 | Discussion [More] | Paul Borrill | Zoom Recap Transcript Chat | Eliminating Timeouts and Retries [Full Summary] |
| #34 OAE #13 | Apr-30, 2025 | Plenary Discussion [More] | Paul Borrill | Zoom Recap Sahas Slides Transcript | Slice Acknowledgements [Full Summary] |
| #33 TAR #9 | Apr-28, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Transcript Chat | Packet Forwarding and Flow Control [Full Summary] |
| #32 CGN #6 | Apr-24, 2025 | Discussion [More] | Paul Borrill | Zoom Recap Transcript Chat | Low Latency in HFT [Full Summary] |
| #31 OAE #12 | Apr-23, 2025 | Plenary Discussion [More] | Paul Borrill | Zoom Recap Requirements Doc Transcript Chat | Requirements Doc Review [Full Summary] |
| #30 TAR #8 | Apr-21, 2025 | Topology & Routing [More] | Paul Borrill | Zoom Recap Transcript Chat | Hardware vs Virtual topologies [Full Summary] |
| #29 CGN #5 | Apr-17, 2025 | Discussion [More] | Paul Borrill | Zoom Recap | Low Latency in HFT [Full Summary] |
| #28 OAE #11 | Apr-16, 2025 | Plenary Discussion [More] | Paul Borrill | Zoom Recap Plenary Slides Visuals Alan’s Slide Transcript Chat | Dead Air & Link Design [Full Summary] |
| #27 TAR #7 | Apr-14, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Transcript Chat | Ant & Bee Algorithms [Full Summary] |
| #26 CGN #4 | Apr-10, 2025 | Discussion [More] | Paul Borrill | Zoom Recap Transcript | PRFAQ Requirements and Ideas [Full Summary] |
| #25 OAE #10 | Apr-09, 2025 | Plenary Discussion [More] | Paul Borrill | Zoom Recap Slides Transcript Chat | Amazon-Style PRFAQ for Atomic Ethernet [Full Summary] |
| #24 TAR #6 | Apr-07, 2025 | Topology & Routing [More] | Alan Karp | Zoom Recap Slide Transcript Chat | Node Addressing vs Port Addressing [Full Summary] |
| #23 CGN #3 | Apr-03, 2025 | Discussion [More] | Paul Borrill | Zoom Recap Transcript | Customer Interview: Factory IoT, Energy IoT [Full Summary] |
| #22 OAE #9 | Apr-02, 2025 | Toward Demand Aware and Self Adjusting Datacenter Networks [More] | Stefan Schmid | Zoom Recap Slides Transcript Chat | Plenary Meeting on Demand-Aware/Self-Adjusting networks [Full Summary] |
| #21 TAR #5 | Mar-31, 2025 | Topology & Routing Workstream Kickoff [More] | Alan Karp | Zoom Recap Transcript Chat | Defining the scope of Topology and Routing in Open Atomic Ethernet [Full Summary] |
| #20 CGN #2 | Mar-27, 2025 | Discussion [More] Transcript | Paul Borrill | Zoom Recap | Advancing network design and hardware for scalable, efficient distributed systems using graph theory, FPGAs, and Cognitive networks. [Full Summary] |
| #19 OAE #8 | Mar-26, 2025 | Physical vs. Software Access [More] | Paul Borrill | Zoom Recap Transcript | Plenary Discussion on Physical vs. Software Access [Full Summary] |
| #18 SAR #4 | Mar-24, 2025 | Workstream Sendoff [More] | Alan Karp | Zoom Recap Transcript Chat | Closing the SAR workstream, and creating the Topology and Routing workstream [Full Summary] |
| #17 OAE #8 | Mar-24, 2025 | Measuring Performance in OpenSHMEM [More] | Camille Coti | Zoom Recap Transcript | Measuring Performance in OpenSHMEM [Full Summary] |
| #16 CGN #1 | Mar-20, 2025 | Deeper Implementation Discussion [More] | Paul Borrill | Zoom Recap Transcript | Defining the Cognitive Networking Discussion Charter [Full Summary] |
| #15 OAE #7 | Mar-19, 2025 | Learnings from Previous Protocol Attempts [More] | Bijan Nowroozi | Transcript | Bijan Naouzi, CTO of OCP, uses this session to revisit Token Ring and ATM as exemplars of deterministic media access, virtual circuits, and fine-grained QoS, contrasting their guarantees with Ethernet’s fundamentally best-effort, collision-prone semantics. The discussion explores which of these “lost” mechanisms—token-based ordering for short in-rack meshes, ATM-style lane disaggregation, feedback-based congestion control, and guaranteed bit/frame rates—should be standardized or left to implementation as we rethink Ethernet for meter-scale atomic links, AI fabrics, and UEC/UA-link era systems. [Full Summary] |
| #14 SAR #3 | Mar-17, 2025 | FPGA Implementation [More] | Alan Karp | Chat | Resilience Continued [Full Summary] |
| #13 FPGA #5 | Mar-13, 2025 | FPGA Implementation [More] | Edmund Humenberger | Transcript | Programmability and Virtualization [Full Summary] |
| #12 OAE #6 | Mar-12, 2025 | Latency vs Bandwidth cont. [More] | Paul | Slides | Analysis of foundation ethernet papers. Followed by a continued discussion on latency vs bandwidth [Full Summary] |
| #11 SAR #2 | Mar-10, 2025 | Workstream Kickoff [More] | Alan Karp | Transcript | Defining the basic security model for OAE [Full Summary] |
| #10 FPGA #4 | Mar-6, 2025 | FPGA Implementation [More] | Edmund Humenberger | discussion of Atomic Ethernet and outlines the current phase of the standardization process. [Full Summary] | |
| #09 OAE #5 | Mar-5, 2025 | Latency vs Bandwidth pt 2 [More] | Paul Borrill | What is the causal graph in Wolfram Physics? Transcript | Causal Graphs and Multiway Systems. Followed by a continued discussion on latency vs bandwidth [Full Summary] |
| #08 SAR #1 | Mar-4, 2025 | Workstream Kickoff [More] | Alan Karp | Transcript | Security & Reliability workstream kickoff. Defining the basic threat model for OAE [Full Summary] |
| #07 FPGA #3 | Feb-27, 2025 | Emulation & Simulation [More] | Edmund Humenberger | Transcript | The group discusses various aspects of FPGA and chip design for network processing. [Full Summary] |
| #06 OAE #4 | Feb-26, 2025 | Large vs Small [More] | Paul Borrill | Slides Transcript Slides(Dugan) | Demonstration of Daedaelus Emulation, Simulation, and TIKTYKTIK guarantees. Also, a discussion on latency vs bandwidth tradeoffs. [Full Summary] |
| #05 FPGA #2 | Feb-20, 2025 | Network on Chips [More] | Edmund Humenberger | Slides | Atomic Ethernet guarantees, software interface. Also discusses NoC opportunities for OAE [Full Summary] |
| #04 OAE #3 | Feb-19, 2025 | Latency vs Bandwidth [More] | Paul Borrill | Slides Transcript | Voting Standards, discussion about Latency vs Bandwidth tradeoffs. [Full Summary] |
| #03 FPGA #1 | Feb-13, 2025 | Workstream Kickoff & Goals [More] | Edmund Humenberger | Slides Transcript | Learnings from years of standardization experience, and what work needs to be done in OAE to ensure success.[Full Summary] |
| #02 OAE #2 | Feb-12, 2025 | Defining Atomicity in Networks [More] | Ken Birman | Slides Transcript | What is an atomic network, and what is required to create a network of atomic transactions?[Full Summary] |
| #01 OAE #1 | Feb-5, 2025 | The Race to 0 [More] | David Lariviere | Slides Transcript | Open Atomic Ethernet’s kickoff session framed OAE as an OCP forum to re-imagine Ethernet around ultra-low latency, explicit knowledge-of-loss, and tightly managed membership at rack scale, with Prof. David Lariviere leading a first deep dive on real-time requirements and design trade-offs. [Full Summary] |
Ethernet History and Interesting Links
https://ethw.org/Milestones:Ethernet_Local_Area_Network_(LAN),_1973-1985
https://ieeemilestones.ethw.org/Milestone-Proposal:Manchester_Code
Interesting Recordings (TAP)
| Index | Date | Topics | Speakers | Slides | Abstract |
|---|---|---|---|---|---|
| TAP #149 | Dec-31, 2025 | From Timeouts to Transactions for Deterministic, Observable Layer-2 Network | Paul Borrill, Ph.D. | Slides | [EPISODE 5 – From Timeouts to Transactions].
Open Atomic Ethernet (OAE) advances a critique of conventional Ethernet and datacenter networking: treating links as probabilistic channels forces correctness and liveness to be reconstructed above Layer 2 via timeout-and-retry, inducing cascading uncertainty, tail-latency amplification, and metastable failure behavior. This talk summarizes OAE’s 2025 progress toward restoring Ethernet as a reliably predictable substrate by redefining each link as a small transactional state machine whose behavior is directly observable and mechanically verifiable. Contributions & scaffolding include:
The presentation further motivates a graph-based Layer-2 topology discovery and “scouting” model for chiplet-scale and datacenter-scale fabrics, reinforcing the explanatory value of resilience and predictability over globally optimal routing. We close with implications for AI/ML training infrastructure — where the dominant constraint shifts from bandwidth to failure ambiguity — and outline a 2026 path toward open governance, iterative adaptation, and standardization transfer. |
| TAP #121 | Jan-01, 2025 | Ethernet Spacetime | Paul Borrill | Slides | THERE IS NO GLOBAL DRUM BEAT. In Episodes 1 through 4 we expressed doubts about the common belief system of a Newtonian view of the world in this community. We showed how to think about race conditions, and why Timeouts and Retries (TAR) are the root of all evil. Our conclusion is that Timestamps are an Illusion. They can’t be fixed by software.
The quest for a single, consistent timeline across distributed systems collides with the reality that physics itself does not provide a universal notion of time—and in quantum mechanics (the machine code of our universe), there is no consistent causal order at all. We cannot, therefore, rely on this illusion of an irreversible drumbeat on an inaccessible “real line” to provide linear time order for events in our networked systems. Although timestamps will remain indispensable in engineering practice, we must recognize them as approximations rather than absolutes, and design our systems accordingly. |
| TAP #118 | Dec-04, 2024 | Timeout and Retry (TAR) in Distributed Systems | Paul Borrill | Slides | [EPISODE 4 – Why we can’t have nice things in Distributed Systems.]
– Instants are meaningless, only intervals (on the same computer/timeline) are relevant – Photons don’t carry timestamps, but timestamps are carried by photons – The speed of light is the “pivot” around which time and space evolve – Timeout and retry (TAR) on different timelines will silently corrupt data structures – Shannon entropy is a logarithm. The logarithm of zero (no information) is minus infinity. – Bayesian approaches require a prior belief, which can be unbounded (zero to infinity). Actually, it’s much worse: can be [-∞,-1,-0,+0,+1,+∞]. We can’t do Bayesian statistics under those conditions, mathematically, their results are undefined – Shannon Entropy is uncertainty, and the same problem applies when you apply the set [-∞,-1,-0,+0,+1,+∞]. to Information and Entropy p*log(p) – Measurements “appear” instantaneous because there is no background of time on which to measure anything. Timestamps don’t help with causal order. |
| TAP #105 | Jul-17, 2024 | Race Conditions and Exactly Once Semantics In Distributed Systems | Paul Borrill | Slides | [EPISODE 3 — How a static PTP hierarchy can be made dynamic to support causal failover for distributed systems].
In Episode 1(What) & Episode 2 (Why) we showed how misunderstandings accumulate within a Newtonian framework of time, and how this leads to lost transactions and corrupted data. In this Episode we help the audience make the leap from Newtonian Time (what we know for certain that just ain’t so) to Post-Newtonian Time (relativistic SR/GR, and QM — Indefinite Causal Order (ICO). – PTP is widely available in Datacenters, we propose experiments to falsify beliefs about Newtonian Time. – All is not lost. The excellent engineering behind PTP and PTM, can still be used with a different perspective, by using the clock hierarchy to build Causal Trees and reliable failover, to help address race conditions and achieve Exactly Once Semantics |
| TAP #102 | Jun-05, 2024 | A Façade of Newtonianism in Networking and it’s Consequences | Paul Borrill | Slides | [EPISODE 2 – Hidden assumptions about causality lead to lost & corrupted data]. When we think about clocks as an incrementing number, we are committing the FITO fallacy — “Forward In-Time-Only” Thinking
– Counterfactuals, i.e., “events that could have occurred but eventually did not, play a unique role in quantum mechanics in that they exert causal effects despite their non-occurrence” – Clock Synchronization Error is indistinguishable from Latency – Irreversibility (Monotonicity) is an illusion not guaranteed by physics, unless we build Ancilla to explicitly manipulate causality – Irreversibility and “causal order” are IN THE EYE OF THE OBSERVER—not guaranteed to be consistent across different observers |
| TAP #98 | Apr-10, 2024 | Clock Coherence: from Terrestrial Microdatacenters to Interstellar Attoprobeswarms | Paul Borrill | Slides | [EPISODE 1 — What] There is no now. You cannot synchronize clocks the way you think. Talk Originally given at the 2023 Asilomar Microcomputer Workshop presented live with Jonathan Gorard.
Motivation: (1) To get people thinking about the nature of time and causality, as far removed from the Earth (and TAI/GPS) as possible. (2) To stimulate “First Principles Thinking” for Distributed Systems. See The Last Theory Newsletter on What is the causal graph in Wolfram Physics
– Clocks can be disseminated, but require interaction to be synchronized – Simultaneity planes don’t exist (except in an empty frozen universe) Einstein proved this over 100 years ago Why do we still think we can synchronize clocks? – Network Time Protocol (NTP) and Precision Time Protocol (PTP) are causal TREES — choose your root, and how you do failover – Entanglement and indefinite causal order are the new relativity (Not restricted to low relative velocities or atomic scales) – We cannot assume spacetime is irreversible and monotonic – Irreversibility and monotonicity is in the Eye of the Observer |
| Index | Date | Topics | Speakers | Slides | Abstract |
|---|---|---|---|---|---|
| TAP #127 | Feb-26, 2025 | Distributed Run-time Environments for HPC | Camille Coti, Ph.D. | Slides | What is a run-time environment? What does it do? Why is it necessary? What are the current challenges? |
| TAP #120 | Dec-18, 2024 | Yours, Mine and Ours | Pat Helland, Daniel May | Slides | Reconciling SETS in O(n log n) Where n = Set-Difference-Size (Not Set-Size) Offers New Opportunities in System Design! |
| TAP #119 | Dec-11, 2024 | Taming Tail Latency: Performance Isolation in Large-Scale Distributed Systems | Jonathan Perry | Slides | Strategies for Mitigating Performance Interference in Cloud-Native Systems |
| TAP #109 | Aug-23, 2024 | Programming Distributed Systems | Mae Milano | Slides | We’ve built a cloud, but we don’t know how to program it yet. |
| TAP #106 | Jul-31, 2024 | What is a clock and what can we do with it? | Pauli Erker | Slides | Explores the thermodynamic and quantum limits of timekeeping, analyzing autonomous quantum clocks, fundamental accuracy-resolution trade-offs, and their implications for quantum computation. |
| TAP #101 | May-22, 2024 | Does Accurate Time Matter for Tasks that need Consensus? | Ken Birman | Slides | Everyone is focused on AI and ML. So Cornell took the next step and created Cascade: a new runtime environment for time-sensitive machine learning and AI inference. |
| TAP #100 | May-08, 2024 | PTP: There at the Beginning | John Eidson, Kang Lee, Doug Arnold | Slides | Fireside chat with experts on PTP and time sync |
| TAP #95 | Feb-28, 2024 | Reliable, Fast Failure Detection with Deterministic Interactions in Datacenters | Davide Rovelli | Slides | Presents a system-driven approach to reliable failure detection in distributed systems by leveraging synchronous interactions, minimizing network and processing jitter, and integrating OS-level monitoring to improve fault tolerance and performance. |
| TAP #85 | Oct-11, 2023 | ÜberNIC: an Ethernet Adapter with CXL and PTM Support | Seth Friedman | Slides | The ÜberNIC presentation showcases a high-performance FPGA-based network controller with SmartNIC capabilities, ultra-low latency, extensive protocol support, and advanced networking features designed for modern data centers. |
| TAP #80 | Aug-23, 2023 | Precision Time in the Last Centimeters with PCIe PTM: A Deeper Dive | Kevin Stanton | Slides | Intel’s PCIe PTM technology enables precise nanosecond-level time synchronization between network cards and CPUs for critical applications. |
| TAP #66 | Feb-15, 2023 | Trading off Consistency and Availability in Cyber-Physical Systems | Edward A. Lee | Slides | CAL theorem: as network latency increases, either inconsistency or unavailability or both must increase. How can you design software systems that are deterministic. |
| Papers We Love Too / San Francisco | Jul-14, 2016 | Paul Borrill: Lamport’s Unfinished Revolution | Paul Borrill | Slides | A review of Leslie Lamport’s 1978 paper “Time, Clocks, and the Ordering of Events in a Distributed System,” framing it as an unfinished revolution. The talk explores the tensions between computational and physical concepts of time, the illusion of simultaneity, and introduces concepts like reversible computing and quantum entanglement as pathways to completing the revolution for a general theory of concurrency. It argues that failures are transcendentally related to the notion of time and that computer scientists and physicists have used two different languages to articulate this problem. |
| Papers We Love Too / San Francisco | Jul-14, 2016 | Time, Clocks, and the Reordering of Events | Paul Borrill and Adrian Cockcroft — Communicating Sequential Papers | Slides | In a “Papers We Love” talk, Paul Borrill revisits Leslie Lamport’s seminal 1978 paper on distributed systems, calling it an “unfinished revolution”. He praises Lamport’s “happened before” relation for establishing event ordering based on causality rather than a universal clock. Borrill critiques the “Forward-In-Time-Only” (FITO) mindset in computer science, which leads to cascading failures from timeouts and retries, contrasting it with the flexible nature of time in modern physics, such as relativity and quantum entanglement. He advocates for “reversible computing,” where local operations can be “unhappened” or “unsent” to handle failures deterministically without system-wide rollbacks. Borrill concludes by calling for a new “general theory of concurrency” that unifies computer science and physics to complete Lamport’s revolution. |