I/O Channel Abstraction for Accelerator Device Kernels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Accelerator device kernels in data centers are limited in their ability to communicate and cooperate with a large percentage of other accelerator devices due to their restricted communication faculties, often being confined to specific physical communication paths, which hinders the utilization of disaggregated resources.
Innovation Solution
The implementation of a system that provides I/O channel abstraction for accelerator device kernels, allowing them to establish logical communication paths across various physical communication paths, optimizing communication capacity and quality of service based on availability data and quality of service parameters, enabling cooperation with a broader range of devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If accelerator device kernels are configured to establish connections with other accelerator device kernels through physical communication paths, then communication capability is improved, but device complexity increases and adaptability decreases
Solution Approach 1:
The patent introduces a communication abstraction layer that acts as an intermediary between the kernel and physical communication paths. This abstraction layer handles the complexity of establishing connections and managing communication protocols, allowing kernels to communicate with diverse accelerator devices without requiring complex configuration. The abstraction layer translates high-level communication requests into specific protocol implementations, thereby reducing kernel configuration complexity while maintaining communication capability.
Solution Approach 2:
The patent creates a universal communication interface that enables accelerator device kernels to communicate with multiple types of accelerator devices through different physical communication paths (PCIe, CXL, NVLink, etc.). This universal interface handles protocol translation and adaptation, allowing a single kernel to work with various device types without requiring separate configuration for each communication path, thus improving adaptability without proportionally increasing complexity.
2Adaptability or versatility
If accelerator device kernels are limited to specific physical communication paths, then device complexity is reduced, but adaptability and resource utilization worsen
Solution Approach 1:
The communication abstraction layer serves as a mediator that manages multiple physical communication paths behind a unified interface. When a kernel needs to access resources on another accelerator device, the abstraction layer automatically selects the appropriate physical communication path based on availability, performance requirements, and current system state. This shields kernels from the complexity of managing multiple communication paths while enabling broad resource accessibility.
Solution Approach 2:
The patent segments the communication functionality into distinct layers: the kernel layer that defines communication needs, the abstraction layer that manages communication paths and protocols, and the physical layer that handles specific communication technologies. This segmentation allows each layer to operate independently, reducing the complexity burden on kernels while enabling access to diverse resources through the coordinated work of lower layers.
3Reliability
If communication protocols are hardcoded in kernel by developer, then communication reliability is improved, but adaptability to different accelerator devices worsens
Solution Approach 1:
The patent implements dynamic protocol selection where the communication abstraction layer determines which protocol to use based on the target device type and communication requirements, rather than having fixed protocols in the kernel. The system can dynamically adapt the communication protocol at runtime, selecting from PCIe, CXL, NVLink, or other protocols as needed. This dynamic approach maintains communication reliability by using appropriate protocols for each device while achieving broad device compatibility.
Solution Approach 2:
The communication abstraction layer changes communication parameters (protocol type, data format, transfer size) based on the target accelerator device and workload requirements. Instead of hardcoding single protocol parameters in the kernel, the system adjusts these parameters dynamically according to the specific communication scenario, ensuring reliable communication with diverse devices while maintaining adaptability to different device types and communication conditions.
Data Source
AI summary
Technologies for providing I/O channel abstraction for accelerator device kernels include an accelerator device comprising circuitry to obtain availability data indicative of an availability of one or more accelerator device kernels in a system, including one or more physical communication paths to each accelerator device kernel. The circuitry is also configured to receive a request to establish a logical communication path between a kernel of the present accelerator device and another accelerator device kernel and establish, in response to the request and as a function of the obtained availability data, the logical communication path between the kernel of the present accelerator device and the other accelerator device kernel.