Disk-based Storage Research Articles

Tremors: Privacy-Breaching Inference of Computing Tasks Using Vibration-Based Condition Monitors

We propose the adaptation of vibration-based condition monitoring systems and techniques, popularly used in industrial condition-based maintenance, for identifying the possibility of compromising the privacy of personal computing systems. This work exploits the automated fan-based heat dissipation features and read/write operations of disk-based storage, commonly present in personal computers, to read computing task-specific vibration signatures on the computer’s cabinet/case. These vibration signatures are then used to identify the broad classes of tasks being executed on a separate computer without ever needing to log into the monitored machine. This work builds upon the premise that heterogeneous tasks have distinct computing requirements, which translates to variations in the amount of heat generated by the computer’s processor, eventually leading to variations in the computer’s heat control fan speed. The variations in the fan’s speed and the frequency of read/write operations to disk-based storage create unique vibration signatures, which maps uniquely to the computer’s processing operations, leading to a breach of privacy of the computer. Our work’s preliminary results suggest that computer-based tasks can be mapped from their vibration signatures with an accuracy of at least <inline-formula><tex-math notation="LaTeX">$70\%$</tex-math></inline-formula> . We additionally study the task identification granularity of such an approach.

Accelerating HDF5 I/O for Exascale Using DAOS

The Hierarchical Data Format 5 (HDF5) has long been defined as one of the most prominent data models, binary file formats and I/O libraries for storing and managing scientific data. Introduced in the late 90s when POSIX I/O was the standard, the library has since then been continuously improved to respond and adapt to the ever-growing demands of high-performance computing (HPC) software and hardware. Given the limitations of POSIX I/O and with the emergence of new technologies such as object stores, non-volatile memory, and SSDs, the need for an interface that can efficiently store and access data at scale through new paradigms has become more and more pressing. The Distributed Asynchronous Object Storage (DAOS) file system is an emerging file system that aims at responding to those demands by taking disk-based storage out of the loop. We present in this article the research efforts that have been taking place to prepare the HDF5 library for Exascale using DAOS. By enabling and defining a new storage file format, we focus on the benefits that it delivers to the applications in terms of features and performance.

Transparent Asynchronous Parallel I/O Using Background Threads

Moving toward exascale computing, the size of data stored and accessed by applications is ever increasing. However, traditional disk-based storage has not seen improvements that keep up with the explosion of data volume or the speed of processors. Multiple levels of non-volatile storage devices are being added to handle bursty I/O, however, moving data across the storage hierarchy can take longer than the data generation or analysis. Asynchronous I/O can reduce the impact of I/O latency as it allows applications to schedule I/O early and to check their status later. I/O is thus overlapped with application communication or computation or both, effectively hiding some or all of the I/O latency. POSIX and MPI-I/O provide asynchronous read and write operations, but lack the support for non-data operations such as file open and close. Users also have to manually manage data dependencies and use low-level byte offsets, which requires significant effort and expertise to adopt. In this article, we present an asynchronous I/O framework that supports all types of I/O operations, manages data dependencies transparently and automatically, provides implicit and explicit modes for application flexibility, and error information retrieval. We implemented these techniques in HDF5. Our evaluation of several benchmarks and application workloads demonstrates it effectiveness on hiding the I/O cost from the application.

Viper

Key-value stores (KVSs) have found wide application in modern software systems. For persistence, their data resides in slow secondary storage, which requires KVSs to employ various techniques to increase their read and write performance from and to the underlying medium. Emerging persistent memory (PMem) technologies offer data persistence at close-to-DRAM speed, making them a promising alternative to classical disk-based storage. However, simply drop-in replacing existing storage with PMem does not yield good results, as block-based access behaves differently in PMem than on disk and ignores PMem's byte addressability, layout, and unique performance characteristics. In this paper, we propose three PMem-specific access patterns and implement them in a hybrid PMem-DRAM KVS called Viper. We employ a DRAM-based hash index and a PMem-aware storage layout to utilize the random-write speed of DRAM and efficient sequential-write performance PMem. Our evaluation shows that Viper significantly outperforms existing KVSs for core KVS operations while providing full data persistence. Moreover, Viper outperforms existing PMem-only, hybrid, and disk-based KVSs by 4--18X for write workloads, while matching or surpassing their get performance.

The first disk-based custodial storage for the ALICE experiment

We proposed a disk-based custodial storage as an alternative to tape for the ALICE experiment at CERN to preserve its raw data. The proposed storage system relies on Redundant Array of Independent Nodes (RAIN) layout – the implementation of erasure coding in the EOS storage suite, which is developed by CERN – for data protection and takes full advantage of high-density Just-Bunch-Of-Disks (JBOD) enclosures to maximize storage capacity as well as to achieve cost-effectiveness comparable to tape. The system we present provides 18 PB of total raw capacity from the 18 set of high-density JBOD enclosures attached to 9 EOS front-end servers. In order to balance between usable space and data protection, the system will stripe a file into 16 chunks on the 4-parity enabled RAIN layout configured on top of 18 containerized EOS FSTs. Although the reduction rate of available space increases up to 33:3% with this layout, the estimated annual data loss rate drops down to 8:6 × 10−5%. In this paper, we discuss the system architecture of the disk-based custodial storage, 4-parity RAIN layout, deployment automation, and the integration to the ALICE experiment in detail.

An intelligent energy efficient storage system for cloud based big data applications

Storage technology has emerged as an indispensable paradigm for processing various applications in cloud data centers. The storage infrastructure consisting of Hard Disk Drives (HDDs) and Solid-State Drives (SSDs) accounts for high energy consumption. Also, the trade-offs between HDDs and SSDs in terms of cost and energy consumption are extremely high. Therefore, disk-based storage subsystems need to be more energy efficient. This paper proposes an intelligent energy-efficient hybrid disk storage system. The proposed system recognizes the frequently used data from traces of applications. Replica management along with data layout is used to allocate frequently used files in hot disks and the other files in cold disks. The request is executed using an intelligent scheduling technique that searches and selects the disk based on its states. The scheduling technique selects either an idle or active disk without spinning the standby disks. The selection procedure also considers minimum waiting time for the active disk and maximum remaining idle time for the idle disk. The proposed system has been implemented by adding disk management in the cloud environment, which proved to be highly effective in achieving a 39% power savings with an 18.26% decrease in execution time.

Industrial-strength OLTP using main memory and many cores

GaussDB, and its open source version named openGauss, are Huawei's relational database management systems (RDBMS), featuring a primary disk-based storage engine. This paper presents a new storage engine for GaussDB that is optimized for main memory and many cores. We started from a research prototype which exploits the power of the hardware but is not useful for customers. This paper describes the details of turning this prototype to an industrial storage engine, including integration with GaussDB. Standard benchmarks show that the new engine provides more than 2.5x performance improvement to GaussDB for full TPC-C on Intel's x86 many-cores servers, as well as on Huawei TaiShan servers powered by ARM64-based Kunpeng CPUs.

PAP: power aware prediction based framework to reduce disk energy consumption

Disk-based storage subsystems account for a significant portion of the energy consumption in both low and high end servers. Therefore, there is a dire need to reduce the server power consumption of the hard disks. In this work, the power-aware framework has been proposed, which efficiently switches the disk into standby, active and idle states, leading to the least power consumption. Firstly, the trace of a real-world application has been generated and processed. The frequently used queries from the trace have been analyzed and prefetched in SSD cache using the data placement policy which lead to 78.5% cache hits. Subsequently, the idle time threshold policy has been executed, which regularly monitors and compares the disk idle time with its threshold value. Later, the request arrival threshold policy predicts the breakeven time using the ensemble machine learning model, which yields 87% accuracy with 3.5% average error rate. Only upon exceeding the threshold values, the disk would smartly be placed in the standby mode; otherwise, it would remain in the idle state to avoid the high power spins in case of frequent requests. Finally, the experimental results have been validated with the existing benchmarks using SSD as a cache, which leads to 75% average power savings.

Seeking an alternative to tape-based custodial storage

In November 2018, the KISTI Tier-1 centre started a project to design, develop and deploy a disk-based custodial storage with error rate and reliability compatible with a tape-based storage. This project has been conducted in collaboration with KISTI and CERN; especially the initial design was laid out from the intensive discussion with CERN IT and ALICE. The system design of the disk-based custodial storage consisted of high density JBOD enclosures and erasure coding data protection, implemented in EOS, the open-source storage management developed at CERN. In order to balance the system reliability, data security and I/O performance, we investigated the possible SAS connections of JBOD enclosures to the front-end node managed by EOS and the technology constraints of interconnections in terms of throughput to accommodate large number of disks foreseen in the storage. This project will be completed and enter production before the start of LHC Run3 in 2021. In this paper we present the detailed description on the initial system design, the brief results of test equipment for the procurement, the deployment of the system, and the further plans for the project.

ExaHDF5: Delivering Efficient Parallel I/O on Exascale Computing Systems

Scientific applications at exascale generate and analyze massive amounts of data. A critical requirement of these applications is the capability to access and manage this data efficiently on exascale systems. Parallel I/O, the key technology enables moving data between compute nodes and storage, faces monumental challenges from new applications, memory, and storage architectures considered in the designs of exascale systems. As the storage hierarchy is expanding to include node-local persistent memory, burst buffers, etc., as well as disk-based storage, data movement among these layers must be efficient. Parallel I/O libraries of the future should be capable of handling file sizes of many terabytes and beyond. In this paper, we describe new capabilities we have developed in Hierarchical Data Format version 5 (HDF5), the most popular parallel I/O library for scientific applications. HDF5 is one of the most used libraries at the leadership computing facilities for performing parallel I/O on existing HPC systems. The state-of-the-art features we describe include: Virtual Object Layer (VOL), Data Elevator, asynchronous I/O, full-featured single-writer and multiple-reader (Full SWMR), and parallel querying. In this paper, we introduce these features, their implementations, and the performance and feature benefits to applications and other libraries.

Research on Simulation Optimization of Disk-based Storage Operation Mode in Warehouse

Research on Simulation Optimization of Disk-based Storage Operation Mode in Warehouse

A light-weight log-based hybrid storage system

The Internet of Things (IoT) and cloud computing are two important technologies in our life. They are integrated together to support each other. However, with the rapid development of IoT, the disk-based storage systems in clouds fail to process and analyze data timely from IoT devices. In order to improve the storage capabilities of traditional disk-based storage systems, we present a light-weight log-based hybrid storage system, which comprises of the phase change memory (PCM), flash memory-based SSD and disks. These devices are merged as a single block device with a continuous linear-address space. We also design and implement a log-based hybrid file system called HFS for the block device. By combining the semantic characteristics of data with distinct performance characteristics of devices, HFS optimizes the file system layout and overcomes the shortcomings of traditional log-structured systems. We implement HFS in the Linux kernel and compare it with modern file systems such as Ext4 and F2FS. Evaluation results show the efficiency of the proposed light-weight log-based hybrid storage system.

A method and tool to recover data deleted from a MongoDB

DBMS stores an important data, which is one of the important analytical subjects for analysis in digital forensics. The technique of recovering deleted data from the DBMS plays an important role in finding the evidence in forensic investigation cases. Although relational DBMS is used as important data storage until now, NoSQL DBMSs is used more often due to the growing pursue of Big Data. This increases the potential to analyze a NoSQL DMBS in forensic cases. In reality, data from approximately 26,000 servers has been deleted by a massive ransom attack on vulnerable MongoDB server. Therefore, investigation of internal structure analysis and deleted data recovery techniques of NoSQL DBMS is essential.In this paper, we research the recovery method on deleted data in MongoDB that is widely used. We have analyzed the internal structures of the WiredTiger and MMAPv1 storage engines, which are the MongoDB's disk-based storage engines. Moreover, we have implemented the recovery algorithm as a tool as well as have evaluated its performance on real and self-generated experiment data.

Energy-Aware Disk Storage Management : Online Approach with Application in DBMS

Energy consumption has become a first-class optimization goal in design and implementation of data-intensive computing systems. This is particularly true in the design of database management systems (DBMS), which was found to be the major consumer of energy in the software stack of modern data centers. Among all database components, the storage system is one of the most power-hungry elements. In previous work, dynamic power management (DPM) techniques that make real-time decisions to transition the disks to low-power modes are normally used to save energy in storage systems. In this paper, we tackle the limitations of DPM proposals in previous contributions. We introduced a DPM optimization model integrated with model predictive control (MPC) strategy to minimize power consumption of the disk-based storage system while satisfying given performance requirements. It dynamically determines the state of disks and plans for inter-disk data fragment migration to achieve desirable balance between power consumption and query response time. Via analyzing our optimization model to identify structural properties of optimal solutions, we propose a fast-solution heuristic DPM algorithm that can be integrated in large-scale disk storage systems for efficient state configuration and data migration. We evaluate our proposed ideas by running simulations using extensive set of synthetic workloads based on popular TPC benchmarks. Our results show that our solution significantly outperforms the best existing algorithm in both energy savings and response time.

Me-CLOCK:A Memory-Efficient Framework to Implement Replacement Policies for Large Caches

Solid State Drives (SSDs) have been extensively deployed as the cache of hard disk-based storage systems. The SSD-based cache generally supplies ultra-large capacity, whereas managing so large a cache introduces excessive memory overhead, which in turn makes the SSD-based cache neither cost-effective nor energy-efficient. This work targets to reduce the memory overhead introduced by the replacement policy of SSD-based cache. Traditionally, data structures involved in cache replacement policy reside in main memory. While these in-memory data structures are not suitable for SSD-based cache any more since the cache is much larger than ever. We propose a memory-efficient framework which keeps most data structures in SSD while just leaving the memory-efficient data structure (i.e., a new bloom proposed in this work) in main memory. Our framework can be used to implement any LRU-based replacement policies under negligible memory overhead. We evaluate our proposals via theoretical analysis and prototype implementation. Experimental results demonstrate that, our framework is practical to implement most replacement policies for large caches, and is able to reduce the memory overhead by about $10 \times$ .

Coral: A Cloud-Backed Frugal File System

With simple access interfaces and flexible billing models, cloud storage has become an attractive solution to simplify the storage management for both enterprises and individual users. However, traditional file systems with extensive optimizations for local disk-based storage backend can not fully exploit the inherent features of the cloud to obtain desirable performance. In this paper, we present the design, implementation, and evaluation of Coral, a cloud based file system that strikes a balance between performance and monetary cost. Unlike previous studies that treat cloud storage as just a normal backend of existing networked file systems, Coral is designed to address several key issues in optimizing cloud-based file systems such as the data layout, block management, and billing model. With carefully designed data structures and algorithms, such as identifying semantically correlated data blocks, kd-tree based caching policy with self-adaptive thrashing prevention, effective data layout, and optimal garbage collection, Coral achieves good performance and cost savings under various workloads as demonstrated by extensive evaluations.

Blurred Persistence

Persistent memory provides data durability in main memory and enables memory-level storage systems. To ensure consistency of such storage systems, memory writes need to be transactional and are carefully moved across the boundary between the volatile CPU cache and the persistent main memory. Unfortunately, cache management in the CPU cache is hardware-controlled. Legacy transaction mechanisms, which are designed for disk-based storage systems, are inefficient in ordered data persistence of transactions in persistent memory. In this article, we propose the Blurred Persistence mechanism to reduce the transaction overhead of persistent memory by blurring the volatility-persistence boundary. Blurred Persistence consists of two techniques. First, Execution in Log executes a transaction in the log to eliminate duplicated data copies for execution. It allows persistence of the volatile uncommitted data, which are detectable with reorganized log structure. Second, Volatile Checkpoint with Bulk Persistence allows the committed data to aggressively stay volatile by leveraging the data durability in the log, as long as the commit order across threads is kept. By doing so, it reduces the frequency of forced persistence and improves cache efficiency. Evaluations show that our mechanism improves system performance by 56.3% to 143.7% for a variety of workloads.

LoneStar RAID

The need for huge storage archives rises with the ever growing creation of data. With today’s big data and data analytics applications, some of these huge archives become active in the sense that all stored data can be accessed at any time. Running and evolving these archives is a constant tradeoff between performance, capacity, and price. We present the LoneStar RAID, a disk-based storage architecture, which focuses on high reliability, low energy consumption, and cheap reads. It is designed for MAID systems with up to hundreds of disk drives per server and is optimized for “write once, read sometimes” workloads. We use dedicated data and parity disks, and export the data disks as individually accessible buckets. By intertwining disk groups into a two-dimensional RAID and improving single-disk reliability with intradisk redundancy, the system achieves an elastic fault tolerance that can at least recover all 3-disk failures. Furthermore, we integrate a cache to offload parity updates and a journal to track the RAID’s state. The LoneStar RAID scheme provides a mean time to data loss (MTTDL) that competes with today’s erasure codes and is optimized to require only a minimal set of running disk drives.

Disk failure prediction model for storage systems based on disk SMART technology

Recent years have seen remarkable advancements in storage systems designed to carry data. The data reliability issue has become increasingly important, as certain adverse factors can impact storage systems and cause data loss. In this study, we developed a disk failure prediction model using disk SMART technology for disk-based storage systems. The proposed model works in a twofold process: first, a parameter reliability model is established based on the actual threshold and attribute values of the SMART parameters, (which we determined through theoretical discussion, then experimentation and analysis) and second, a disk reliability model is built according to the parameter reliability model and parameter weights as determined. The disk reliability failure threshold is then determined by analyzing the SMART parameter data and parameter weights in tandem with the disk reliability model to fully determine the level of disk reliability. This approach eventually allows the user to migrate low-level reliability disks to backup disks, thereby improving the overall reliability of the storage system. Experimental results proved that the proposed approach is feasible and effective.

Toward Scalable Indexing for Top-&lt;inline-formula&gt; &lt;tex-math notation="TeX"&gt;\(k\) &lt;/tex-math&gt;&lt;/inline-formula&gt; Queries

A top- k query retrieves the best \(k\) tuples by assigning scores for each tuple in a target relation with respect to a user-specific scoring function. This paper studies the problem of constructing an indexing structure for supporting top- k queries over varying scoring functions and retrieval sizes. The existing research efforts can be categorized into three approaches: list- , layer- , and view-based approaches. In this paper, we mainly focus on the layer-based approach that pre-materializes tuples into consecutive multiple layers. We first propose a dual-resolution layer that consists of coarse-level and fine-level layers. Specifically, we build coarse-level layers using skylines , and divide each coarse-level layer into fine-level sublayers using convex skylines . To make our proposed dual-resolution layer scalable , we then address the following optimization directions: 1) index construction; 2) disk-based storage scheme; 3) the design of the virtual layer; and 4) index maintenance for tuple updates. Our evaluation results show that our proposed method is more scalable than the state-of-the-art methods.