Global Shared Memory Notification via Host Fabric Interface
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multi-processor data processing systems face scalability limitations due to bandwidth constraints, particularly in system memory access, and lack efficient support for tasks to share address spaces across physical or logical partitions in distributed environments, hindering the productive scaling of parallel computing across multiple nodes.
Innovation Solution
A method and system that utilize a host fabric interface (HFI) to enable local notification of completion for global shared memory (GSM) operations across multiple nodes, allowing tasks to execute GSM operations and monitor their completion, thereby facilitating efficient parallel job execution by managing effective addresses and preventing unauthorized access to memory spaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If SMP architecture with shared system memory is used, then inter-processor communication is simple, but scalability is limited due to bandwidth constraints
Solution Approach 1:
The system divides the monolithic shared memory architecture into distributed memory segments across multiple nodes. Each node has its own local memory, and the system provides a global address space that maps to physical memory locations across different nodes, enabling scalable parallel processing without bandwidth bottlenecks.
Solution Approach 2:
The patent introduces a global address space dimension that overlays the physical distributed memory architecture. This allows processors to access memory using unified addressing while the underlying system scales across multiple nodes with non-uniform memory access patterns.
2Productivity
If NUMA architecture with distributed memory is used, then system scalability is improved, but access latency becomes non-uniform
Solution Approach 1:
The system provides a universal global address space that can address memory locations regardless of which node they reside on. This unified addressing mechanism simplifies programming while the underlying hardware handles non-uniform access patterns, maintaining scalability benefits.
Solution Approach 2:
The patent introduces a global address space translation mechanism that acts as an intermediary between processor requests and physical memory locations. This layer manages the complexity of non-uniform access by providing unified addressing, allowing the system to scale while managing latency through intelligent address mapping.
3Reliability
If tasks execute on separate nodes without shared address space, then node independence is maintained, but parallel job execution efficiency decreases
Solution Approach 1:
The system merges the address spaces of multiple tasks into a unified global address space while maintaining physical separation of memory resources. This allows tasks on different nodes to efficiently share data and cooperate on parallel jobs, as they can reference the same logical addresses even though physical memory resides on different nodes.
Data Source
AI summary
In a global shared memory (GSM) environment, a method provides local notification of completion of a global shared memory (GSM) operation processed by a first task executing at a local node of the distributed system. The system includes multiple nodes on which different tasks of a single job execute and perform GSM operations that are received from a second task via a via host fabric interface (HFI) and associated HFR window assigned to the first tasks. The local task initiates execution of a GSM operation on the local node. The task then monitors for and detects a completion of the execution of the GSM operation on the local node. When the task detects completion of the execution of the GSM operation, the task issues an internal notification to inform the locally-executing tasks of the completion of the GSM operation.