Search

The High Energy Photon Source (HEPS) is expected to produce a substantial volume of data, lead to immense data I/O pressure during computing. Inefficient data I/O can significantly impact computing performance. To address this challenge, firstly, we have developed a data I/O framework for HEPS. This framework consists of three layers: data channel layer, distributed memory management layer, and I/O interface layer. It mask the underlying data differences in formats and sources, while implementing efficient I/O methods. Additionally, it supports both stream computing and batch computing. Secondly, we have designed a data processing pipeline scheme aimed at reducing I/O latency and optimizing I/O bandwidth utilization during the processing of high-throughput data. This involves breaking down the computing task into several stages, including data loading, data pre-processing, data processing, and data writing, which are executed asynchronously and in parallel. Finally, we introduce the design of stream data I/O process. The primary objective of stream data I/O is to enable real-time online processing of raw data, avoiding I/O bottlenecks caused by disk storage. This approach ensures the stability of data transmission and integrates distributed memory management to guarantee data integrity in memory.

Large-scale high-energy physics experiments generate scientific data at the scale of petabytes or even exabytes, requiring high-performance data IO for processing. However, in large computing centers, computing and storage devices are typically separated. Large-scale data transfer has become a bottleneck for some data-intensive computing tasks, such as data encoding and decoding, compression, sorting, etc. The time spent on data transfer can account for 50% of the entire computing task. The larger the amount of data accessed, the more significant this cost becomes. One attractive solution to address this problem is to offload a portion of data processing to the storage layer. However, modifying traditional storage systems to support computation offloading is often cumbersome and requires a broad understanding of their internal principles. Therefore, we have designed a flexible software framework called XkitS, which builds a computable storage system by extending the existing storage system EOS. This framework is deployed on the EOS FTS storage server and offloads computational tasks by invoking the computing capabilities (CPU, FPGA, etc.) on FTS. Currently, it has been tested and applied in the data processing of the Large High Altitude Air Shower Observatory (LHAASO), and the results show that the time spent on data decoding using the computable storage technology is half of that using the original method.

With the advancement of many large HEP experiments, the amount of data that needs to be processed and stored has increased significantly, so we must upgrade computing resources and improve the performance of storage software. This article discusses porting the EOS software from the x86_64 architecture to the aarch64 architecture, with the aim of finding a more cost-effective storage solution. In the process of porting, the biggest challenge is that many dependent packages do not have aarch64 version and need to be compiled by ourselves, and the assembly part of the software code also needs to be adjusted accordingly. Despite these challenges, we have successfully ported the EOS code to the aarch64. This article discusses the current status and plans for the software port as well as performance testing after porting.

Synchrotron radiation sources (SRS) produce a huge amount of image data. This scientific data, which needs to be stored and transferred losslessly, will bring great pressure on storage and bandwidth. The SRS images have the characteristics of high frame rate and high resolution, and traditional image lossless compression methods can only save up to 30% in size. Focus on this problem, we propose a lossless compression method for SRS images based on deep learning. First, we use the difference algorithm to reduce the linear correlation within the image sequence. Then we propose a reversible truncated mapping method to reduce the range of the pixel value distribution. Thirdly, we train a deep learning model to learn the nonlinear relationship within the image sequence. Finally, we use the probability distribution predicted by the deep leaning model combined with arithmetic coding to fulfil lossless compression. Test result based on SRS images shows that our method can further decrease 20% of the data size compared to PNG, JPEG2000 and FLIF.

In this paper we will describe the LHAASO distributed computing system based on virtualization and cloud computing technologies. Particularly, we discuss the key points of integrating distributed resources. A solution of integrating cross-domain resources is proposed, which adopt the Openstack+HTCondor to make the distributed resources work as a whole resource pool. A flexible resource scheduling strategy and a job scheduling policy are presented to realize the resource expansion on demand and the efficient job scheduling to remote sites transparently, so as to improve the overall resource utilization. We will also introduce the deployment of the computing system located in Daocheng, the LHAASO observation base using cloud-based architecture, which greatly helps to reduce the operation and maintenance cost as well as to make sure the system availability and stability. Finally, we will show running status of the system.

High energy physics (HEP) experiments, such as LHAASO, produce a large amount of data, which is usually stored and processed on distributed sites. Nowadays, the distributed data management system faces some challenges such as global file namespace, efficient data access and storage. Focusing on those problems, this paper proposed a cross-domain data access file system (CDFS), applying data deduplication and compression as the storage-optimized engine, aiming at dynamically building an aggregate view of multiple distributed storages and accessing data in a fast and efficient way. The test based on the raw data of LHAASO experiment showed that the CDFS could present a unique repository based on distributed sites in LHAASO. And the storage-optimized engine reduces the storage consumption of the raw data by more than 50%.

High-energy physics computing is a typical data-intensive calculation. Each year, petabytes of data needs to be analyzed, and data access performance is increasingly demanding. The tiered storage system scheme for building a unified namespace has been widely adopted. Generally, data is stored on storage devices with different performances and different prices according to different access frequency. When the heat of the data changes, the data is then migrated to the appropriate storage tier. At present, heuristic algorithms based on artificial experience are widely used in data heat prediction. Due to the differences in computing models of different users, the accuracy of prediction is low. A method for predicting future access popularity based on file access characteristics with the help of LSTM deep learning algorithm is proposed as the basis for data migration in hierarchical storage. This paper uses the real data of high-energy physics experiment LHAASO as an example for comparative testing. The results show that under the same test conditions, the model has higher prediction accuracy and stronger applicability than existing prediction models.

Beijing Spectrometer (BESIII) experiment has produced hundreds of billions of events. The traditional event-wise accessing of BESIII Offline Software System is not effective for the selective accessing with low rate during a physics analysis. In this paper, an event-based data management system (EventDB) is introduced, which can effectively alleviate the problems of low efficiency of data processing and low utilization of resources. Firstly, an indexing system based on NoSQL database is designed. By extracting specified attributes of events, the events interested to the physicists are selected and stored into the database, whilst the real data of event is still stored in ROOT files. For those hot events, the real event data can also be cached into EventDB to improve the access performance. The data analysis workflow of HEP experiments is needed to change if the EventDB system is applied. The analysis program queries the corresponding event index from database, then get event data from database if the event is cached, or get data from ROOT files if it is not cached. Finally, the test on more than one hundred billion physics events shows the query speed was greatly improved over traditional file-based data management systems.