算网融合DPU可定制化能力分层方法

Layered Method of DPU Customization Capability for Computing and Network Convergence

社会数字化转型推动着行业提升“联接+计算”能力,促进网络向算网融合发展。数据处理单元(Data Processing Unit,DPU)是在海量数据高效传输和处理需求下快速发展起来的新技术形态,部署位置处于网络和计算的衔接点,是算网融合的关键技术。文章首先阐述DPU的典型应用场景和方案,其次分析DPU的组成架构和可编程资源,最后提出DPU的可定制化能力分层方法,用于DPU的软硬件资源和可定制化能力细分,有助于用户根据业务需求和研发能力进行产品评估和选型、二次开发和业务优化,有利于DPU在算网融合场景中发挥其灵活和高效的数据处理能力实现算网业深度协同,助力DPU产业生态发展和算网融合技术创新。

The societal digital transformation is compelling various industries to augment their"connectivity+computing"capabilities,propelling network development towards the fusion of computing and networking.In response to the burgeoning need for processing and transmitting massive data,the Data Processing Unit(DPU)has rapidly evolved as a new technological paradigm.Positioned at the pivotal intersection of networking and computation,the DPU emerges as a fundamental technology supporting computing and network convergence.Firstly,the DPU's typical application scenarios are outlined,and then its architectural framework and programmable features are analysed.Finally,a layered method of DPU's customization capabilities is proposed.This method allows for refined segmentation of the DPU's programmable resources and customization capabilities,assists users in product evaluation and selection,secondary development,and business optimization according to their specific operational needs and R&D capabilities.Furthermore,it contributes to a deeper synergy between computing and networking enabled by DPU's flexible and efficient data processing ability,and supports the growth of the DPU industry ecosystem and innovation in computing and network convergence technology.