基于虚拟化的软件定义DSP计算技术研究

Research on Software Defined DSP Computing Technology Based on Virtualization

随着天基计算密集型任务快速增长,星上任务对计算的需求也随之迅猛上涨。星载计算机系统通常采用异构电子系统,其“算力”被封闭在独立的星载计算机内,各计算资源不能彼此共享,容易产生资源浪费,且各个嵌入式软件和可编程逻辑采用非标准的编程,不支持统一升级和综合调用。针对此问题,以异构CPU和DSP计算资源作为研究重点,设计一种基于虚拟化技术的软件定义DSP计算架构,通过分析当前所面临的抽象化和多租户的技术挑战,并借鉴异构GPU和FPGA相关虚拟化技术,综合运用硬件抽象层、时分复用虚拟化、接口调用重定向、组件注册和数据路由等技术,实现DSP异构计算资源多用户共享和高效统一调用。试验验证表明,软件定义DSP计算架构可实现星载异构计算系统软硬件灵活解耦,并支持多个用户和应用同时访问和调用单个DSP计算资源。

With the rapid growth of space-based computationally intensive tasks,the demand for computing in onboard tasks has also skyrocketed.Spaceborne computer systems typically use heterogeneous electronic systems,with their"computing power"enclosed in independent onboard computers,where computing resources cannot be shared with each other,leading to resource waste.Additionally,various embedded software and programmable logic use non-standard programming,which does not support unified upgrades and comprehensive calls.To address this issue,a software defined DSP computing architecture based on virtualization technology is designed with heterogeneous CPU and DSP computing resources as the research focus.By analyzing the current challenges of abstraction and multi tenant technology,and drawing on virtualization technologies related to heterogeneous GPUs and FPGAs,technologies such as hardware abstraction layer,time division multiplexing virtualization,interface call redirection,component registration,and data routing are comprehensively utilized,the multi user sharing and efficient unified invocation of DSP heterogeneous computing resources is realized.Experimental verification shows that software defined DSP computing architecture can achieve flexible decoupling of software and hardware in spaceborne heterogeneous computing systems,and support multiple users and applications accessing and calling a single DSP computing resource simultaneously.