Introduction: On October 20, at the 2021 Hangzhou Cloud Conference, Ali Cloud released the 4th Generation DpCA architecture, which is upgraded to the new eRMDA network architecture, which is the industry’s first large-scale flexible RDMA acceleration capability.

On October 20, alibaba Cloud unveiled the fourth-generation DpCA architecture at the 2021 Hangzhou Cloud Conference. Compared to the previous generation, the fourth generation of DpCA has improved the basic performance, including 300% more storage IOPS, 100% more network PPS and 80% less network latency. Meanwhile, it has been upgraded to the new eRDMA network architecture, which is the industry’s first large-scale flexible RDMA acceleration capability.

Storage IOPS up to 3 million, network PPS up to 50 million, the minimum network delay of 5 microseconds, ali Cloud fourth generation Of Dragon architecture, a number of indicators, again refresh the industry’s highest level. Based on the leading basic performance, the fourth generation Of Shenlong will bring further performance improvement for common applications such as database, AI and big data: MySQL database performance increase by 60%, and Nginx SSL scenarios up to 420%.

The Divine Dragon architecture is a new generation of hardware and software integrated virtualization technology developed by Ali Cloud. It unloads virtualization on dedicated chips for processing and implements hardware acceleration IO engine. With ultra-high performance and flexibility, it is the best carrier of cloud native. In terms of security, the latest generation of DpCA is equipped with enterprise-level security chips and is equipped with trusted computing and encryption computing capabilities to achieve the system’s trusted tamper-proof and data availability invisible. Among them, the fourth generation of DpCA architecture unique flexible RDMA acceleration capability, will benefit the majority of Internet applications.

Elastic RDMA, or eRDMA for short, is China’s first large-scale RDMA acceleration capability on the cloud, which can greatly improve the communication efficiency of large-scale distributed computing, and can dynamically expand with the cluster size of Ali Cloud, which can easily build the largest RDMA distributed computing network in China.

With the exponential expansion of Internet data volume, the scale of distributed computing in data centers becomes larger and larger, and the cost of mass data transmission between large-scale clusters becomes larger and larger. Compared with traditional TCP networks, elastic RDMA can greatly reduce the network communication delay caused by large-scale cluster network interconnection, greatly improve the performance and cost performance of distributed computing, and help build distributed computing applications with larger scale and larger data volume on the cloud.

In AI deep learning scenarios, eRDMA can improve the training performance of large-scale distributed NLP and visual computing by 30%. In the Spark big data scenario, the Spark computing performance of large-scale distributed big data is improved by 30%. In the scenario of Redis KV database, the maximum throughput of Redis mixed read and write is improved by 130%.

ERDMA enables RDMA network to support general computing scenarios from popular AI and HPC applications, which will bring computing architecture innovation and enable the great development of cloud native technologies such as Microservice, Serverless and Service Mesh.

Since its launch in 2017, Aliyun cloud servers are all based on the Self-developed Shenlong architecture, supporting a variety of elastic computing product forms such as elastic bare metal, virtual machine and elastic container instances, and serving multiple industries such as medical care, new government affairs, intelligent manufacturing, Internet and education. Jiang Jiangwei, head of the basic product division of Ali Cloud, said that based on the DCP architecture, Ali Cloud will continue to promote the universal benefits of high-quality computing power and continue to provide a driving force for social innovation.

The original link

This article is the original content of Aliyun and shall not be reproduced without permission.