A GPU-Accelerated In-Memory Metadata Management Scheme forLarge-Scale Parallel File Systems

Driven by the increasing requirements of high-performance computing applications,supercomputers are prone to containing more and more computing nodes.Applications running on such a large-scale computing system are likely to spawn millions of parallel processes,which usually generate a burst of I/O requests,introducing a great challenge into the metadata management of underlying parallel file systems.The traditional method used to overcome such a challenge is adopting multiple metadata servers in the scale-out manner,which will inevitably confront with serious network and consistence problems.This work instead pursues to enhance the metadata performance in the scale-up manner.Specifically,we propose to improve the performance of each individual metadata server by employing GPU to handle metadata requests in parallel.Our proposal designs a novel metadata server architecture,which employs CPU to interact with file system clients,while offloading the computing tasks about metadata into GPU.To take full advantages of the parallelism existing in GPU,we redesign the in-memory data structure for the name space of file systems.The new data structure can perfectly fit to the memory architecture of GPU,and thus helps to exploit the large number of parallel threads within GPU to serve the bursty metadata requests concurrently.We implement a prototype based on BeeGFS and conduct extensive experiments to evaluate our proposal,and the experimental results demonstrate that our GPU-based solution outperforms the CPU-based scheme by more than 50%under typical metadata operations.The superiority is strengthened further on high concurrent scenarios,e.g.,the high-performance computing systems supporting millions of parallel threads.

Adaptive and scalable load balancing for metadata server cluster in cloud-scale file systems

Big data is an emerging term in the storage indus- try, and it is data analytics on big storage, i.e., Cloud-scale storage. In Cloud-scale （or EB-scale） file systems, load bal- ancing in request workloads across a metadata server cluster is critical for avoiding performance bottlenecks and improv- ing quality of services. Many good approaches have been pro- posed for load balancing in distributed file systems. Some of them pay attention to global namespace balancing, making metadata distribution across metadata servers as uniform as possible. However, they do not work well in skew request dis- tributions, which impair load balancing but simultaneously increase the effectiveness of caching and replication, in this paper, we propose Cloud Cache （C2）, an adaptive and scal- able load balancing scheme for metadata server cluster in EB-scale file systems. It combines adaptive cache diffusion and replication scheme to cope with the request load balanc- ing problem, and it can be integrated into existing distributed metadata management approaches to efficiently improve their load balancing performance. C2 runs as follows： 1） to run adaptive cache diffusion first, if a node is overloaded, load- shedding will be used; otherwise, load-stealing will be used; and 2） to run adaptive replication scheme second, if there is a very popular metadata item （or at least two items） causing a node be overloaded, adaptive replication scheme will be used,in which the very popular item is not split into several nodes using adaptive cache diffusion because of its knapsack prop- erty. By conducting performance evaluation in trace-driven simulations, experimental results demonstrate the efficiency and scalability of C2.

ONFS： a hierarchical hybrid file system based on memory, SSD, and HDD for high performance computers

With supercomputers developing towards exascale, the number of compute cores increases dramatically, making more complex and larger-scale applications possible. The input/output （I/O） requirements of large-scale applications, workflow applications, and their checkpointing include substantial bandwidth and an extremely low latency, posing a serious challenge to high performance computing （HPC） storage systems. Current hard disk drive （HDD） based underlying storage systems are becoming more and more incompetent to meet the requirements of next-generation exascale supercomputers. To rise to the challenge, we propose a hierarchical hybrid storage system, on-line and near-line file system （ONFS）. It leverages dynamic random access memory （DRAM） and solid state drive （SSD） in compute nodes, and HDD in storage servers to build a three-level storage system in a unified namespace. It supports portable operating system interface （POSIX） semantics, and provides high bandwidth, low latency, and huge storage capacity. In this paper, we present the technical details on distributed metadata management, the strategy of memory borrow and return, data consistency, parallel access control, and mechanisms guiding downward and upward migration in ONFS. We implement an ONFS prototype on the TH-1A supercomputer, and conduct experiments to test its I/O performance and scalability. The results show that the bandwidths of single-thread and multi-thread ＇read＇/＇write＇ are 6-fold and 5-fold better than HDD-based Lustre, respectively. The I/O bandwidth of data-intensive applications in ONFS can be 6.35 timcs that in Lustre.

Construction and application of LHAASO data processing platform

Purpose The LHAASO project collects trillions of cosmic ray events every year,generating about 10 PB of raw data annually,which brings big challenges for data processing platform.Method The LHAASO data processing platform is built to handle such a large amount of data,which is composed of some subsystems such as data transfer,data storage,high throughput computing and metadata management.Results and conclusions The platform was under construction since 2018 and has been working well since 2021.In this paper,the details of the design,implementation and performance of the data processing platform are presented.

A Ceph-based storage strategy for big gridded remote sensing data

When using distributed storage systems to store gridded remote sensing data in large,distributed clusters,most solutions utilize big table index storage strategies.However,in practice,the performance of big table index storage strategies degrades as scenarios become more complex,and the reasons for this phenomenon are analyzed in this paper.To improve the read and write performance of distributed gridded data storage,this paper proposes a storage strategy based on Ceph software.The strategy encapsulates remote sensing images in the form of objects through a metadata management strategy to achieve the spatiotemporal retrieval of gridded data,finding the cluster location of gridded data through hash-like calculations.The method can effectively achieve spatial operation support in the clustered database and at the same time enable fast random read and write of the gridded data.Random write and spatial query experiments proved the feasibility,effectiveness,and stability of this strategy.The experiments prove that the method has higher stability than,and that the average query time is 38%lower than that for,the large table index storage strategy,which greatly improves the storage and query efficiency of gridded images.