A large model multi-directory scheduling bottom core architecture

By constructing a large-scale multi-directory scheduling underlying core architecture, the instability problem of the scheduling process in the existing system is solved, the stability of directory access and the reliability of instruction execution are achieved, and stable underlying support is provided.

CN122155289APending Publication Date: 2026-06-05陈立波

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
陈立波
Filing Date
2026-03-22
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

The existing large-scale multi-directory scheduling system lacks a generalized underlying core architecture, which leads to directory access deadlocks and instruction execution anomalies during the scheduling process, and cannot provide stable and general underlying support.

Method used

A core architecture for large-scale multi-directory scheduling is constructed, which includes core basic scheduling components and basic operation guarantee mechanisms, including a multi-directory storage module, a scheduling control module, a routing and addressing module, and a directory interlocking module. Combined with an exception handling mechanism, the stability and reliability of the scheduling process are ensured.

Benefits of technology

It achieves stability and reliability in large-scale multi-directory scheduling, avoids access deadlock, ensures the normal execution of scheduling instructions, and provides stable underlying support.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The application discloses a large model multi-directory scheduling bottom layer core architecture, which is provided with a core basic scheduling module, interlocking release and a basic exception bottom mechanism, and can provide general and stable bottom support for large model multi-directory scheduling and is suitable for various large model computing power scheduling bottom layer architecture building scenes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of large model computing power scheduling technology, specifically to a multi-directory scheduling underlying core architecture applied to large model scenarios. Background Technology

[0003] The existing large-scale multi-directory scheduling system lacks a generalized underlying core architecture, and the scheduling process is prone to problems such as directory access deadlock and instruction execution exceptions, which cannot provide stable and general underlying support for large-scale multi-directory scheduling. Summary of the Invention

[0005] This invention provides a core underlying architecture for large-scale model multi-directory scheduling. By combining core basic scheduling components with basic operation guarantee mechanisms, a generalized and stable underlying support architecture for large-scale model multi-directory scheduling is constructed, which is adapted to the basic operation requirements of various large-scale model multi-directory scheduling. Detailed Implementation

[0007] Example

[0008] This architecture is applied to basic scenarios of large-scale model computing power scheduling. The multi-directory storage module stores multiple sets of directory data corresponding to the large-scale model scheduling; the scheduling control module uniformly sends and receives scheduling instructions and distributes them; the routing and addressing module completes data addressing and matching between multiple directories based on the scheduling instructions; the directory interlocking module manages the parallel access behavior of multiple directories, and starts the interlocking release mechanism after the access is completed to avoid access deadlock; the scheduling execution module parses and executes the scheduling instructions, and at the same time handles basic execution exceptions through a basic exception fallback mechanism to ensure that the architecture can continuously complete the basic scheduling operation of large-scale model multi-directory.

Claims

1. A core architecture for large-scale multi-directory scheduling, characterized in that: It includes a multi-directory storage module, a scheduling control module, a routing addressing module, a directory interlocking module, and a scheduling execution module; The directory interlock module is configured with an interlock release mechanism, and the scheduling execution module is configured with a basic exception fallback mechanism.

2. The architecture according to claim 1, characterized in that, It also includes a directory data verification module, which is used to verify the legality of directory data within the multi-directory storage module.

3. The architecture according to claim 1, characterized in that, It also includes a directory status awareness module, which is used to monitor the working status of each directory in the multi-directory storage module in real time.

4. The architecture according to claim 1, characterized in that, It also includes a timing synchronization module, which is used to achieve timing alignment for multi-directory scheduling.

5. The architecture according to claim 1, characterized in that, The scheduling control module has a built-in instruction cache unit for temporarily storing continuous scheduling instructions.

6. The architecture according to claim 1, characterized in that, The multi-directory storage module adopts an independent physical partition structure.

7. The architecture according to claim 1, characterized in that, The modules are connected via bus communication.