Omnixtend Boot Protocol and Coherent Scaleout
summary
OmniXtend defines a race-free boot protocol that configures coherent multi-node RISC-V systems over raw Ethernet frames so a central management server (or a programmable switch) sends device tree payloads to each endpoint, which then set up OmniXtend channels, enable tiling endpoints in a deadlock-avoiding order and either join a single shared-OS NUMA space or boot independent OS instances that share memory via a kernel driver. A prototype uses an x86 management host attached to the switch and claims scalability to hundreds of nodes.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:37Omnixtend Protocol
- 1:25Single Operating System Model
- 2:57Boot Process
- 4:58Management Server
- 6:41Device Tree
- 7:21Sequence Diagram
- 10:04Conclusion
related talks
22mAn Open and Coherent Memory Centric Architecture Enabled by RISC V
31mUnshackling Memory!
23mBreaking the limits of x86
11mXerxes: Extensive Exploration of Scalable Hardware Systems with CXL-Based Simulation...
20mSemperOS: A Distributed Capability System
26mErlang on Xen: Redefining the Software Stack
56mMeet All Scenarios OS: A Distributed O.S. with Feet on the Ground
37mOnyx: Distributed Computing for Clojure