With the development of computing, storage and networking technologies, the way data is stored is changing. For business applications that require large amounts of data, the server’s built-in storage space, or the built-in disk, is no longer sufficient to meet the storage needs.

Therefore, in addition to the built-in storage, the server needs external storage to expand the storage space, such as centralized storage and software-defined storage. The former refers to the way of connecting several disk cabinets through one or several fixed disk array controllers. The disk array controller is used as the data access entrance to provide data access for the application server. The latter is to deploy all the storage-related access and control work in the form of software on a number of ordinary servers, and the application can access the data on these external servers through the network.

Compared with the former, software-defined storage (SDR) has been accepted by more and more industries and applications due to its better horizontal scaling ability, decoupling of software and hardware, higher reliability and availability. At present, there are many domestic manufacturers engaged in software-defined storage, including Yan Rong technology. Yan Rong Technology was founded in 2016, focusing on the research and development of software-defined storage technology. It has made great achievements in a few years. For example, in 2020, Yan Rong Technology not only obtained the first overseas customer of software-defined storage in China, but also achieved a customer growth of nearly 400% and annual sales growth of more than 300%.

So, how does Yan Rong technology achieve such a rapid development? YRCloudFile plays a big part in this.

01 Distributed storage product – YRCloudFile

YrCloudFile is an unstructured data unified storage platform launched by Yan Rong Technology. It can provide customers with high-performance, high-expansion and cloud-native unstructured data storage system in the public cloud and private cloud environment. Shortly after its launch, it won 6th place in the IO500 World Storage Performance Challenge.

Li Junhong, product director of Yan Rong Technology, said that YRCloudFile is a distributed storage product launched after continuous enterprise/user interviews, technical exchanges, comprehensive testing and version iteration. Its core feature is the high performance characteristics of the product. With this feature, YRCloudFile can be a training link in the field of artificial intelligence and automatic driving. And the data loading operation stage in the high computing environment to provide high-speed, stable data access ability.

Figure: YRCloudFile product architecture diagram

Li Junhong told reporters that YRCloudFile has three main features:

High performance under massive small files, using metadata node clustering, multi-level intelligent cache, intelligent pre-read and other technologies, in-depth optimization of massive file read and write performance, directory access hot spots and other issues;

Cloud native container storage provides standard CSI interface, supports PV quota, RESIZE and QoS, and isolates and plans storage resources from capacity and IO performance respectively. Through PV Insight, administrators can view the internal data distribution and hot and cold conditions of storage volumes.

Supporting public cloud/hybrid cloud, YRCloudFile can achieve one-click deployment on public clouds such as Ali cloud, Tencent cloud and AWS, provide high-performance file storage services on public clouds, and make data flow between different levels through the function of cold and hot data layering, thus reducing the total cost of ownership of storage.

Due to its excellent performance, YRCloudFile has been applied in ecological environment by MORE than one hundred customers, including China Mobile, iFlytek, Tucson Future, Neolithic, MORE VFX, Shuken Internet, Bank of China and other leading enterprises in the industry, involving artificial intelligence, automatic driving, film rendering, IoT, finance and other application scenarios.

Yrcloudfile provides high performance service

In the field of autonomous driving, Yan Rong has reached cooperation with Tucson Future, Neolithic and other manufacturers to provide data storage services.

Li Junhong told reporters that autonomous driving belongs to the vertical subdivision field of artificial intelligence, including collection, labeling, calculation and other links. Only through these links can a model be formed to ensure the landing of autonomous driving. However, as more and more sensors such as radar and camera are deployed on the vehicle, the workload of each link is increasing day by day. In particular, some high-performance autonomous driving vehicles have a huge demand for data storage. The amount of data generated can reach about 8TB a day.

Figure: Automatic driving development cycle structure based on YRCloudFile high performance distributed file storage

Therefore, how to efficiently and stably ensure that a large amount of data collected in the process of autonomous driving can quickly form a computational model of autonomous driving has become an important issue of concern to CTO of autonomous driving enterprises. YrCloudFile distributed storage product launched by Yan Rong Technology can solve the problem of data access efficiency in the training process.

According to Li Junhong, the current training process in the field of autonomous driving is facing the bottleneck problem of data access performance. “Self-driving cars drive the sensors will collect the data, to stay after the data collection, deposit will be in various forms to the data center, when the server to get data, due to the storage system to provide the access speed can’t keep up with the GPU speed of processing data, leading to the GPU there idle phenomenon, thus affecting the whole automation training process and efficiency, It causes a waste of GPU resources, and the positioning of YRCloudFile is to solve the problem of data storage and use in the process of automatic driving training “.

Li Junhong believes that the main role of YRCloudFile is to give full play to the computing power of the GPU, in a short time to provide enough data for the GPU to fully calculate, “for example, the use of Yan Rong technology YRCloudFile storage platform, every second can provide access to 5GB to 10GB data for the GPU, It greatly improves the efficiency of GPU to access data. In the field of automatic driving, the final formation of the automatic driving function must be through a lot of training. The function of YRCloudFile is to reduce the training time of each round in a large number of training, so that the time of the self-driving model to market can be advanced. This is the core value that YRCloudFile can bring to the self-driving enterprises.” “Mr Lai said.

Future directions for data storage

YRCloudFile’s distributed storage technology has been recognized by a large number of enterprises, including government, finance, education, healthcare and other organizations. According to Li Junhong, compared with traditional storage methods, the core advantage of distributed storage technology is that it solves two major challenges in current data storage.

That is, through distributed storage software, the concurrent storage access in the cloud platform is distributed to each node of the distributed cluster to meet the storage access performance requirements of the client on the cloud platform; The distributed software can be extended horizontally, so that the capacity of a single cluster can be expanded on demand, and the upper limit of the capacity of a single set of traditional storage devices can be solved. These two advantages make distributed storage technology get a lot of applications and development in cloud environment and large data scale scenarios.

For distributed storage technology, Li Chunhong expressed optimism, and indicated that it is likely to be the mainstream development direction of future storage technology. According to him, the future direction of storage technology will be towards performance fit and scenario-optimized breakthroughs.

“First of all, the future storage technology will definitely adapt to more new components, including brand, network, architecture, etc., which means that the products need to adapt to more excellent data processing capacity, so how to play the performance of the new components, is the upper storage technology suppliers must make improvements. Secondly, from the perspective of development, the future storage technology can only be optimized for some vertical scenarios in order to better layout the scenario. For example, the current autonomous driving, artificial intelligence, video rendering and other fields, only from the perspective of storage to understand, aiming at the characteristics of the field to optimize, can better development.”

At this stage, Yan Rong Technology is in a period of rapid development. In the future, in order to constantly adapt to the needs of various industries for Data storage, Yan Rong Technology will continue to carry out product iteration, expand container applications, hybrid cloud storage, unified unstructured Data management, Data Lakes and other scene applications, and accelerate the coverage of the dominant scene. At the same time, further strengthen the internationalization, strengthen the promotion of the software subscription model and maintain three times the sales growth.