Abstract We establish two types of difference-in-difference (DID) models to study the effects of the policy shocks of the new national system of science and technology on the rationalization and advancedization of industrial structure of national innovation center (NIC) cities. Meanwhile, considering the differences in resource endowment between NIC cities and general cities, we use propensity matching analysis (PSM) to establish matching data on the basis of benchmark data to empirically demonstrate the effect of this policy shock on the industrial structure of NIC cities. The study shows that the policy shock has a promoting effect on the level of industrial structure rationalization in the NIC cities, but not significant effect on the industrial structure advancedization, the improvement of the financing environment can promote the development of industrial structure advancement, the growth of personal wealth has a promoting effect on the industrial structure rationalization, and the increase of the government investment is not conducive to the advancement of industrial structure, although it can promote the rationalization of the industrial structure in China’s NIC cities. Compared with the eastern NIC cities, this policy shock has a greater effect on the rationalization of industrial structure in the Central-western NIC cities of China. In addition, the financing environment and per capita income of NIC cities have a moderating effect on the rationalization of industrial structure under the policy shock. Therefore, NIC cities should actively promote the promotion effect of this policy on the rationalization of industrial structure.

Skip list (skiplist) is a competitive index structure that offers superior concurrency and excellent performance but with high memory overhead and low access locality. Emerging persistent memory (PM) technologies present an opportunity to mitigate the capacity constraint of DRAM. However, data consistency on PM typically results in excessive write overhead. In addition, fast concurrent access to an index is critical to the throughput on high-end contemporary computer systems. In this article, we propose a Partitioned HierArchical SkiplisT calledPhaST, which can simultaneously reduce the skiplist height and improve its access locality, through its hierarchy of component structures, while enabling fast parallel recovery in case of failure. To ensure high concurrency and fast data consistency, we also have developed writelock-free concurrent insert and log-free atomic split. Furthermore, we have developed a durable lock-free concurrent search that can discern transient structural inconsistencies and deliver highly concurrent read operations. We have conducted an extensive evaluation ofPhaSTcompared to state-of-the-art studies such as NV-Skiplist, wB+-Tree, FPTree, and FAST-FAIR. Our evaluation results showPhaSToutperforms other indexing structures by up to 4.05× and 2.87× in single-threaded inserts and searches, and 1.56× and 2.62× in concurrent inserts and searches.

On large-scale high performance computing (HPC) systems, applications are provisioned with aggregated resources to meet their peak demands for brief periods. This results in resource underutilization because application requirements vary a lot during execution. This problem is particularly pronounced for deep learning applications that are running on leadership HPC systems with a large pool of burst buffers in the form of flash or non-volatile memory (NVM) devices. In this paper, we examine the I/O patterns of deep neural networks and reveal their critical need of loading many small samples randomly for successful training. We have designed a specialized Deep Learning File System (DLFS) that provides a thin set of APIs. Particularly, we design the metadata management of DLFS through an in-memory tree-based sample directory and its file services through the user-level SPDK protocol that can disaggregate the capabilities of NVM Express (NVMe) devices to parallel training tasks. Our experimental results show that DLFS can dramatically improve the throughput of training for deep neural networks on NVMe over Fabric, compared with the kernel-based Ext4 file system. Furthermore, DLFS achieves efficient user-level storage disaggregation with very little CPU utilization.

As the Energy Wall and the Reliability Wall become unavoidable, it is a demanding and challenging task to reduce energy consumption in large-scale storage systems in modern data centers while retaining acceptable systems reliability. We propose a reliable energy-efficient storage system called DuoFS, which aims at balancing the energy efficiency, the reliability and the performance of parallel storage systems by seamlessly integrating one HDD-based file system and one SSD-based file system. At the heart of the DuoFS is a transformative middleware layer that dispatches files to the one of the two independent parallel file systems based on the files' I/O access popularity. By replicating popular files to the SSD-based file system and pushing the HDD-based file system into the low-power mode under light workload conditions, DuoFS can reduce significant energy consumption, avoid major factors that harm the storage systems reliability, and extract SSDs good I/O performance. Experimental results show that the DuoFS system saves up to 40% of energy, achieves up to 50% better I/O performance while only sacrificing less than 15% of the system's reliability.

As issues of the Energy Wall and the Reliability Wall become unavoidable, it is a demanding and challenging task to reduce energy consumption in large-scale storage systems in modern data centres while retaining acceptable systems reliability. Most energy conservation techniques inevitably have adverse impacts on the parallel disk systems. To address the reliability issues of energy-efficient parallel storage systems, we propose a reliable energy-efficient storage system called DuoFS, which aims at improving both energy efficiency and reliability of parallel storage systems by seamlessly integrating HDDs and SSDs. With the help of the middleware layer, DuoFS can distribute popular data to SSD-based nodes and put HDD-based nodes into the low-power mode under light workload conditions without modification of the parallel systems.