Refine
Document Type
- Conference proceeding (38)
- Journal article (5)
- Book chapter (5)
Has full text
- yes (48)
Is part of the Bibliography
- yes (48)
Institute
- Informatik (48)
Publisher
Near-data processing in database systems on native computational storage under HTAP workloads
(2022)
Today’s Hybrid Transactional and Analytical Processing (HTAP) systems, tackle the ever-growing data in combination with a mixture of transactional and analytical workloads. While optimizing for aspects such as data freshness and performance isolation, they build on the traditional data-to-code principle and may trigger massive cold data transfers that impair the overall performance and scalability. Firstly, in this paper we show that Near-Data Processing (NDP) naturally fits in the HTAP design space. Secondly, we propose an NDP database architecture, allowing transactionally consistent in-situ executions of analytical operations in HTAP settings. We evaluate the proposed architecture in state-of-the-art key/value-stores and multi-versioned DBMS. In contrast to traditional setups, our approach yields robust, resource- and cost-effcient performance.
Database management systems and K/V-Stores operate on updatable datasets – massively exceeding the size of available main memory. Tree-based K/V storage management structures became particularly popular in storage engines. B+ -Trees [1, 4] allow constant search performance, however write-heavy workloads yield in inefficient write patterns to secondary storage devices and poor performance characteristics. LSM-Trees [16, 23] overcome this issue by horizontal partitioning fractions of data – small enough to fully reside in main memory, but require frequent maintenance to sustain search performance.
Firstly, we propose Multi-Version Partitioned BTrees (MV-PBT) as sole storage and index management structure in key-sorted storage engines like K/V-Stores. Secondly, we compare MV-PBT against LSM-Trees. The logical horizontal partitioning in MV-PBT allows leveraging recent advances in modern B+ -Tree techniques in a small transparent and memory resident portion of the structure. Structural properties sustain steady read performance, yielding efficient write patterns and reducing write amplification.
We integrated MV-PBT in the WiredTiger [15] KV storage engine. MV-PBT offers an up to 2× increased steady throughput in comparison to LSM-Trees and several orders of magnitude in comparison to B+ -Trees in a YCSB [5] workload.
Massive data transfers in modern data-intensive systems resulting from low data-locality and data-to-code system design hurt their performance and scalability. Near-Data processing (NDP) and a shift to code-to-data designs may represent a viable solution as packaging combinations of storage and compute elements on the same device has become feasible. The shift towards NDP system architectures calls for revision of established principles. Abstractions such as data formats and layouts typically spread multiple layers in traditional DBMS, the way they are processed is encapsulated within these layers of abstraction. The NDP-style processing requires an explicit definition of cross-layer data formats and accessors to ensure in-situ executions optimally utilizing the properties of the underlying NDP storage and compute elements. In this paper, we make the case for such data format definitions and investigate the performance benefits under RocksDB and the COSMOS hardware platform.
Real Time Charging (RTC) applications that reside in the telecommunications domain have the need for extremely fast database transactions. Today´s providers rely mostly on in-memory databases for this kind of information processing. A flexible and modular benchmark suite specifically designed for this domain provides a valuable framework to test the performance of different DB candidates. Besides a data and a load generator, the suite also includes decoupled database connectors and use case components for convenient customization and extension. Such easily produced test results can be used as guidance for choosing a subset of candidates for further tuning/testing and finally evaluating the database most suited to the chosen use cases. This is why our benchmark suite can be of value for choosing databases for RTC use cases.
In the present tutorial we perform a cross-cut analysis of database systems from the perspective of modern storage technology, namely Flash memory. We argue that neither the design of modern DBMS, nor the architecture of flash storage technologies are aligned with each other. The result is needlessly suboptimal DBMS performance and inefficient flash utilisation as well as low flash storage endurance and reliability. We showcase new DBMS approaches with improved algorithms and leaner architectures, designed to leverage the properties of modern storage technologies. We cover the area of transaction management and multi-versioning, putting a special emphasis on: (i) version organisation models and invalidation mechanisms in multi-versioning DBMS; (ii) Flash storage management especially on append-based storage in tuple granularity; (iii) Flash-friendly buffer management; as well as (iv) improvements in the searching and indexing models. Furthermore, we present our NoFTL approach to native Flash access that integrates parts of the flash-management functionality into the DBMS yielding significant performance increase and simplification of the I/O stack. In addition, we cover the basics of building large Flash storage for DBMS and revisit some of the RAID techniques and principles.
Flash SSDs are omnipresent as database storage. HDD replacement is seamless since Flash SSDs implement the same legacy hardware and software interfaces to enable backward compatibility. Yet, the price paid is high as backward compatibility masks the native behaviour, incurs significant complexity and decreases I/O performance, making it non-robust and unpredictable. Flash SSDs are black-boxes. Although DBMS have ample mechanisms to control hardware directly and utilize the performance potential of Flash memory, the legacy interfaces and black-box architecture of Flash devices prevent them from doing so.
In this paper we demonstrate NoFTL, an approach that enables native Flash access and integrates parts of the Flashmanagement functionality into the DBMS yielding significant performance increase and simplification of the I/O stack. NoFTL is implemented on real hardware based on the OpenSSD research platform. The contributions of this paper include: (i) a description of the NoFTL native Flash storage architecture; (ii) its integration in Shore-MT and (iii) performance evaluation of NoFTL on a real Flash SSD and on an on-line data-driven Flash emulator under TPCB, C,E and H workloads. The performance evaluation results indicate an improvement of at least 2.4x on real hardware over conventional Flash storage; as well as better utilisation of native Flash parallelism.
Asymmetric read/write storage technologies such as Flash are becoming
a dominant trend in modern database systems. They introduce
hardware characteristics and properties which are fundamentally
different from those of traditional storage technologies such
as HDDs.
Multi-Versioning Database Management Systems (MV-DBMSs)
and Log-based Storage Managers (LbSMs) are concepts that can
effectively address the properties of these storage technologies but
are designed for the characteristics of legacy hardware. A critical
component of MV-DBMSs is the invalidation model: commonly,
transactional timestamps are assigned to the old and the new version,
resulting in two independent (physical) update operations.
Those entail multiple random writes as well as in-place updates,
sub-optimal for new storage technologies both in terms of performance
and endurance. Traditional page-append LbSM approaches
alleviate random writes and immediate in-place updates, hence reducing
the negative impact of Flash read/write asymmetry. Nevertheless,
they entail significant mapping overhead, leading to write
amplification.
In this work we present an approach called Snapshot Isolation
Append Storage Chains (SIAS-Chains) that employs a combination
of multi-versioning, append storage management in tuple granularity
and novel singly-linked (chain-like) version organization.
SIAS-Chains features: simplified buffer management, multi-version
indexing and introduces read/write optimizations to data placement
on modern storage media. SIAS-Chains algorithmically avoids
small in-place updates, caused by in-place invalidation and converts
them into appends. Every modification operation is executed
as an append and recently inserted tuple versions are co-located.
Characteristics of modern computing and storage technologies fundamentally differ from traditional hardware. There is a need to optimally leverage their performance, endurance and energy consumption characteristics. Therefore, existing architectures and algorithms in modern high performance database management systems have to be redesigned and advanced. Multi Version Concurrency Control (MVCC) approaches in data-base management systems maintain multiple physically independent tuple versions. Snapshot isolation approaches enable high parallelism and concurrency in workloads with almost serializable consistency level. Modern hardware technologies benefit from multi-version approaches. Indexing multi-version data on modern hardware is still an open research area. In this paper, we provide a survey of popular multi-version indexing approaches and an extended scope of high performance single-version approaches. An optimal multi-version index structure brings look-up efficiency of tuple versions, which are visible to transactions, and effort on index maintenance in balance for different workloads on modern hardware technologies.
Database management systems (DBMS) are critical performance components in large scale applications under modern update intensive workloads. Additional access paths accelerate look-up performance in DBMS for frequently queried attributes, but the required maintenance slows down update performance. The ubiquitous B+ tree is a commonly used key-indexed access path that is able to support many required functionalities with logarithmic access time to requested records. Modern processing and storage technologies and their characteristics require reconsideration of matured indexing approaches for today's workloads. Partitioned B-trees (PBT) leverage characteristics of modern hardware technologies and complex memory hierarchies as well as high update rates and changes in workloads by maintaining partitions within one single B+-Tree. This paper includes an experimental evaluation of PBTs optimized write pattern and performance improvements. With PBT transactional throughput under TPC-C increases 30%; PBT results in beneficial sequential write patterns even in presence of updates and maintenance operations.
Active storage
(2018)
In brief, Active Storage refers to an architectural hardware and software paradigm, based on collocation storage and compute units. Ideally, it will allow to execute application-defined data ... within the physical data storage. Thus Active Storage seeks to minimize expensive data movement, improving performance, scalability, and resource efficiency. The effective use of Active Storage mandates new architectures, algorithms, interfaces, and development toolchains.
A transaction is a demarcated sequence of application operations, for which the following properties are guaranteed by the underlying transaction processing system (TPS): atomicity, consistency, isolation, and durability (ACID). Transactions are therefore a general abstraction, provided by TPS that simplifies application development by relieving transactional applications from the burden of concurrency and failure handling. Apart from the ACID properties, a TPS must guarantee high and robust performance (high transactional throughput and low response times), high reliability (no data loss, ability to recover last consistent state, fault tolerance), and high availability (infrequent outages, short recovery times).
The architectures and workhorse algorithms of a high-performance TPS are built around the properties of the underlying hardware. The introduction of nonvolatile memories (NVM) as novel storage technology opens an entire new problem space, with the need to revise aspects such as the virtual memory hierarchy, storage management and data placement, access paths, and indexing. NVM are also referred to as storage-class memory (SCM).
The amount of image data has been rising exponentially over the last decades due to numerous trends like social networks, smartphones, automotive, biology, medicine and robotics. Traditionally, file systems are used as storage. Although they are easy to use and can handle large data volumes, they are suboptimal for efficient sequential image processing due to the limitation of data organisation on single images. Database systems and especially column-stores support more stuctured storage and access methods on the raw data level for entiere series.
In this paper we propose definitions of various layouts for an efficient storage of raw image data and metadata in a column store. These schemes are designed to improve the runtime behaviour of image processing operations. We present a tool called column-store Image Processing Toolbox (cIPT) allowing to easily combine the data layouts and operations for different image processing scenarios.
The experimental evaluation of a classification task on a real world image dataset indicates a performance increase of up to 15x on a column store compared to a traditional row-store (PostgreSQL) while the space consumption is reduced 7x. With these results cIPT provides the basis for a future mature database feature.
In this paper we present our work in progress on revisiting traditional DBMS mechanisms to manage space on native Flash and how it is administered by the DBA. Our observations and initial results show that: the standard logical database structures can be used for physical organization of data on native Flash; at the same time higher DBMS performance is achieved without incurring extra DBA overhead. Initial experimental evaluation indicates a 20% increase in transactional throughput under TPC-C, by performing intelligent data placement on Flash, less erase operations and thus better Flash longevity.
In the present paper we demonstrate a novel approach to handling small updates on Flash called In-Place Appends (IPA). It allows the DBMS to revisit the traditional write behavior on Flash. Instead of writing whole database pages upon an update in an out-of-place manner on Flash, we transform those small updates into update deltas and append them to a reserved area on the very same physical Flash page. In doing so we utilize the commonly ignored fact that under certain conditions Flash memories can support in-place updates to Flash pages without a preceding erase operation.
The approach was implemented under Shore-MT and evaluated on real hardware. Under standard update-intensive workloads we observed 67% less page invalidations resulting in 80% lower garbage collection overhead, which yields a 45% increase in transactional throughput, while doubling Flash longevity at the same time. The IPA outperforms In-Page Logging (IPL) by more than 50%.
We showcase a Shore-MT based prototype of the above approach, operating on real Flash hardware – the OpenSSD Flash research platform. During the demonstration we allow the users to interact with the system and gain hands on experience of its performance under different demonstration scenarios. These involve various workloads such as TPC-B, TPC-C or TATP.
In the present paper we demonstrate the novel technique to apply the recently proposed approach of In-Place Appends – overwrites on Flash without a prior erase operation. IPA can be applied selectively: only to DB-objects that have frequent and relatively small updates. To do so we couple IPA to the concept of NoFTL regions, allowing the DBA to place update-intensive DB-objects into special IPA-enabled regions. The decision about region configuration can be (semi-)automated by an advisor analyzing DB-log files in the background.
We showcase a Shore-MT based prototype of the above approach, operating on real Flash hardware. During the demonstration we allow the users to interact with the system and gain hands-on experience under different demonstration scenarios.
Under update intensive workloads (TPC, LinkBench) small updates dominate the write behavior, e.g. 70% of all updates change less than 10 bytes across all TPC OLTP workloads. These are typically performed as in-place updates and result in random writes in page-granularity, causing major write-overhead on Flash storage, a write amplification of several hundred times and lower device longevity.
In this paper we propose an approach that transforms those small in-place updates into small update deltas that are appended to the original page. We utilize the commonly ignored fact that modern Flash memories (SLC, MLC, 3D NAND) can handle appends to already programmed physical pages by using various low-level techniques such as ISPP to avoid expensive erases and page migrations. Furthermore, we extend the traditional NSM page-layout with a delta-record area that can absorb those small updates. We propose a scheme to control the write behavior as well as the space allocation and sizing of database pages.
The proposed approach has been implemented under Shore- MT and evaluated on real Flash hardware (OpenSSD) and a Flash emulator. Compared to In-Page Logging it performs up to 62% less reads and writes and up to 74% less erases on a range of workloads. The experimental evaluation indicates: (i) significant reduction of erase operations resulting in twice the longevity of Flash devices under update-intensive workloads; (ii) 15%-60% lower read/write I/O latencies; (iii) up to 45% higher transactional throughput; (iv) 2x to 3x reduction in overall write
amplification.
Introduction to the special issue on self‑managing and hardware‑optimized database systems 2022
(2023)
Data management systems have evolved in terms of functionality, performance characteristics, complexity, and variety during the last 40 years. Particularly, the relational database management systems and the big data systems (e.g., Key-Value stores, Document stores, Graph stores and Graph Computation Systems, Spark, MapReduce/Hadoop, or Data Stream Processing Systems) have evolved with novel additions and extensions. However, the systems administration and tasks have become highly complex and expensive, especially given the simultaneous and rapid hardware evolution in processors, memory, storage, or networking. These developments present new open problems and challenges to data management systems as well as new opportunities.
The SMDB (International Workshop on Self-Managing Database Systems) and HardBD&Active (Joint International Workshop on Big Data Management on Emerging Hardware and Data Management on Virtualized Active Systems) workshops organized in conjunction with the IEEE ICDE (International Conference on Data Engineering) offered two distinct platforms for examining the above system-related challenges from different perspectives. The SMDB workshop looks into developing autonomic or self-* features in database and data management systems to tackle complex administrative tasks, while the HardBD&Active workshop focuses on harnessing hardware technologies to enhance efficiency and performance of data processing and management tasks. As a result of these workshops, we are delighted to present the third special issue of DAPD titled “Self-Managing and Hardware-Optimized Database Systems 2022,” which showcases the best contributions from the SMDB 2021/2022 and HardBD&Active 2021/2022 workshops.
Near-Data Processing (NDP) is a key computing paradigm for reducing the ever growing time and energy costs of data transport versus computations. With their flexibility, FPGAs are an especially suitable compute element for NDP scenarios. Even more promising is the exploitation of novel and future non-volatile memory (NVM) technologies for NDP, which aim to achieve DRAM-like latencies and throughputs, while providing large capacity non-volatile storage.
Experimentation in using FPGAs in such NVM-NDP scenarios has been hindered, though, by the fact that the NVM devices/FPGA boards are still very rare and/or expensive. It thus becomes useful to emulate the access characteristics of current and future NVMs using off-the-shelf DRAMs. If such emulation is sufficiently accurate, the resulting FPGA-based NDP computing elements can be used for actual full-stack hardware/software benchmarking, e.g., when employed to accelerate a database.
For this use, we present NVMulator, an open-source easy-to-use hardware emulation module that can be seamlessly inserted between the NDP processing elements on the FPGA and a conventional DRAM-based memory system. We demonstrate that, with suitable parametrization, the emulated NVM can come very close to the performance characteristics of actual NVM technologies, specifically Intel Optane. We achieve 0.62% and 1.7% accuracy for cache line sized accesses for read and write operations, while utilizing only 0.54% of LUT logic resources on a Xilinx/AMD AU280 UltraScale+ FPGA board. We consider both file-system as well as database access patterns, examining the operation of the RocksDB database when running on real or emulated Optane-technology memories.
Rapidly growing data volumes push today's analytical systems close to the feasible processing limit. Massive parallelism is one possible solution to reduce the computational time of analytical algorithms. However, data transfer becomes a significant bottleneck since it blocks system resources moving data-to-code. Technological advances allow to economically place compute units close to storage and perform data processing operations close to data, minimizing data transfers and increasing scalability. Hence the principle of Near Data Processing (NDP) and the shift towards code-to-data. In the present paper we claim that the development of NDP-system architectures becomes an inevitable task in the future. Analytical DBMS like HPE Vertica have multiple points of impact with major advantages which are presented within this paper.
When forecasting sales figures, not only the sales history but also the future price of a product will influence the sales quantity. At first sight, multivariate time series seem to be the appropriate model for this task. Nontheless, in real life history is not always repeatable, i.e. in the case of sales history there is only one price for a product at a given time. This complicates the design of a multivariate time series. However, for some seasonal or perishable products the price is rather a function of the expiration date than of the sales history. This additional information can help to design a more accurate and causal time series model. The proposed solution uses an univariate time series model but takes the price of a product as a parameter that influences systematically the prediction. The price influence is computed based on historical sales data using correlation analysis and adjustable price ranges to identify products with comparable history. Compared to other techniques this novel approach is easy to compute and allows to preset the price parameter for predictions and simulations. Tests with data from the Data Mining Cup 2012 demonstrate better results than established sophisticated time series methods.