Big Data Workloads Research Articles

DNA computing and meta-heuristic-based algorithm for big data task scheduling in cloud computing

With the advent of cloud computing, there is a need to enhance both the methods and algorithms of big data workloads for task scheduling. Due to the global spread of services with changing task load circumstances and different cloud client demands, big data task scheduling in cloud systems is a time-consuming process. The proposed approach emphasises the necessity for efficient big data task scheduling in the cloud computing, which exacerbate data processing. Virtual machines frequently utilise all three types of physical resources: CPU, memory, and storage. Big data task scheduling is one of the most important implications of cloud computing application resource management, and this research work meticulously offers a task scheduling technique for advancing cloud computing.

OPTIMIZING SQL DATABASES FORBIG DATA WORKLOADS: TECHNIQUES AND BEST PRACTICES

In the era of big data, SQL databases face significant challenges in handling vast volumes of data efficiently. This article explores optimization techniques and best practices for enhancing the performance and scalability of SQL databases in handling big data workloads. The study addresses the significant challenges faced by traditional SQL databases, including scalability issues, performance bottlenecks, resource constraints, and data integration complexities. Through a comprehensive methodology involving literature review, case studies, expert interviews, and performance testing, the research identifies effective strategies such as indexing, partitioning, sharding, and caching. Findings from case studies in e-commerce and financial services sectors demonstrate substantial improvements in query performance and resource utilization, validating the practical benefits of these optimization techniques. The study underscores the importance of a multifaceted approach to database optimization, integrating both theoretical and practical insights to address the complexities of big data environments. By staying informed and adopting the latest optimization strategies, database administrators and IT professionals can ensure their SQL databases remain efficient, scalable, and capable of managing the increasing demands of large-scale data processing, ultimately enabling organizations to derive valuable insights from their data.

Serverless-like platform for container-based YARN clusters

Serverless computing is an emerging paradigm that has gained a lot of relevance in recent years, as it allows users to consume computing resources without worrying about the underlying infrastructure and pay only for what they actually use. Most current services that implement this paradigm typically rely on the Function-as-a-Service (FaaS) model, which works perfectly for simple applications based on stateless functions triggered by specific events. However, these services are not designed to run more complex applications with intricate interactions, usually presenting a significant degree of configuration difficulty and/or low ability to customise the execution environment. They also tend to be designed for short and simple workloads, with some services even limiting their maximum runtime to just a few minutes. In this paper, we present a platform based on Hadoop YARN oriented to the execution of Big Data workloads in a containerised and serverless way, so that the resources allocated to such containers are automatically and dynamically scaled according to their actual usage. An experimental evaluation has been carried out to compare our serverless-like platform with a standard YARN deployment when executing Big Data workloads concurrently. Our results have shown experimental evidence of enhancing both performance and overall resource efficiency, providing runtime reductions and resource usage improvements of up to 41% and 50%, respectively.

An analysis of cache configuration’s impacts on the miss rate of big data applications using gem5

This work aims to analyze the impacts of cache configurations on miss rates of big data benchmarks with varying level 1 instruction (L1I) and data (L1D) caches using the gem5 simulator. The cache miss rate of nine big data applications from four benchmark suits is analyzed with different cache configurations, such as increasing the cache size, varying the associativity, and altering the line size. The gem5 provides a versatile platform for conducting detailed experiments. The study sheds light on the relationship between cache and big data workloads, thus offering insights into optimizing cache configurations? effect on miss rates for improved performance.

Distributed File System to Leverage Data Locality for Large-File Processing

Over the past decade, significant technological advancements have led to a substantial increase in data proliferation. Both scientific computation and Big Data workloads play a central role, manipulating massive data and challenging conventional high-performance computing architectures. Efficiently processing voluminous files using cost-effective hardware remains a persistent challenge, limiting access to new technologies for individuals and organizations capable of higher investments. In response to this challenge, AwareFS, a novel distributed file system, addresses the efficient reading and updating of large files by consistently exploiting data locality on every copy. Its distributed metadata and lock management facilitate sequential and random I/O patterns with minimal data movement over the network. The evaluation of the AwareFS local-write protocol demonstrated efficiency across various update patterns, resulting in a performance improvement of approximately 13%, while benchmark assessments conducted across diverse cluster sizes and configurations underscored the flexibility and scalability of AwareFS. The innovative distributed mechanisms outlined herein are positioned to contribute to the evolution of emerging technologies related to the computation of data stored in large files.

Main memory controller with multiple media technologies for big data workloads

SRAM and DRAM memory technologies have been dominant in the implementations of memory subsystems. In recent years, and mainly driven by the huge memory demands of big data applications, NVRAM technology has emerged as a denser memory technology, enabling the design of new hybrid DRAM/NVRAM memory hierarchies that combine multiple memory media technologies to balance memory capacity, latency, cost, and endurance. Two main approaches are being applied to the design of hybrid memory hierarchies: the hybrid address space approach, which relies on the programmer or the operating system to choose the memory technology where each memory page should be stored; and the (only) NVM address space approach, where a faster technology (e.g. commodity DRAM) is needed to acts as a cache of NVRAM to boost the performance. This approach presents architectural challenges such the organization of metadata (e.g. cache tags) and the selection of the proper technology for each memory component. In contrast to existing approaches, this work proposes a memory controller that leverages novel memory technologies such as eDRAM and MRAM to mitigate NVRAM bus contention and improve the performance of the NVM address space. The devised solution proposes a two-level cache hierarchy in the memory controller: a SRAM sector cache and a (x)RAM cache. The (x)RAM cache, much denser, helps significantly reduce the number of accesses to NVRAM. Experimental results show that implementing the (x)RAM cache with eDRAM or MRAM is the best performing approach. Moreover, the eRAM is able to improve the SRAM cache miss penalty by up to 50% and 80%, and overall system performance by 15% and 23%.

A Mini Tutorial of Processing in Memory: From Principles, Devices to Prototypes

Data movement overheads caused by the recent explosion in big data applications have made traditional von Neumann architecture fails to tackle big data workloads. Processing in Memory (PIM), where computational tasks are performed within the memory, has drawn increasing attention. To present meaningful insights to readers, we divide current PIM paradigm into charge-based and resistance-based categories according to different memory devices. This mini tutorial aims to provide a concise overview of the implementation of PIM schemes, highlighting important macro prototypes in artificial intelligence applications that have been released in the past five years.

Extending an asynchronous runtime system for high throughput applications: A case study

Current supercomputers are mostly composed of vast numbers of nodes enhanced with accelerators (usually in the form of GPUs). However, having these heterogeneous designs in the forefront have exposed the software toolchains and application designers to the underlying complexities of an ever evolving hardware substrate. The need for a more dynamic view from the system software (i.e., compilers and runtimes) has become more apparent in these environments. Due to this, adaptive, fine grain runtime systems have seen a rise in popularity in the past decades. With low overhead and small tasks, these runtimes help to hide long latency operations by exploiting the massive concurrency presented in different application workflows. Such features allow the reduction of idle time (a result from ever deeper and complex memory hierarchies and memory types) with the execution of unrelated work across the machine. Of these runtimes, the Asynchronous Many Task (AMT) Runtimes are excellent exemplars as they can efficiently map onto hardware substrates and exhibit a high degree of latency hiding. Because of their latency tolerant characteristics, applications such as Graph Analytics and Big Data applications (which are latency sensitive) can use these runtimes very efficiently. Thanks to these characteristics, we present how a careful design can help to exploit the properties of an AMT when running high latency applications such as the ones encountered in the Big Data domain. In addition, when combined with introspection / adaptive capabilities, the runtime can further exploit optimization opportunities during its execution based on the ever changing state of the underlying hardware substrate. As a vehicle for this exploration, we use the Performance Open Community Runtime (P-OCR) to test all these concepts with Big Data workloads.

Toward a novel engine for compiler optimization space exploration of big data workloads

AbstractRecently, big data specific technologies have been emerging, including domain‐specific languages, software frameworks, databases, third‐party libraries, and so forth. These techniques are successful in concealing the low‐level details by producing high‐level code, which is passed through the conventional compilation cycle for generating hardware operable code. Several optimization opportunities exist in the compiler which can assist in meeting the processing deadlines of big data workloads, through optimized machine code. However, the existing iterative compilation techniques are not enough for the exploration of big data applications. In this regard, a novel engine has been presented for exploiting the compiler optimization space of big data workloads. The engine is comprised of training and testing phases. During the training stage, the big data application is optimized with Mitigates the Compiler Phase‐ordering (MiCOMP) and genetic algorithm (GA) optimization sequences, which are executed with train datasets. In the testing stage, the test datasets are executed only for the best 300 optimization sequences discovered at the training stage. The proposed engine has been tested with graph mining, machine learning, and text search categories of big data applications using a wide range of real‐world and synthetic datasets. Overall, the engine is , , and faster than Iterative Optimization for the Data Center (IODC), MiCOMP, and GA respectively in exploiting the compiler search space for big data workloads. Further, the integration of best‐10 and best‐3 techniques with the engine brings a speedup of and . The compiler level exploitation of general‐purpose machines incurs no extra overhead, no heavy computing, and no personnel cost. Also, the overall performance of big data specialized software solutions can be enhanced by compiling their high‐level code with suitable compiler optimizations.

RENDA: Resource and Network Aware Data Placement Algorithm for Periodic Workloads in Cloud

The Hadoop enabled cloud platforms are gradually becoming preferred computational environment to execute scientific big data workloads in a periodic manner. However, it is observed that the default data placement approach of such cloud platforms is not the efficient one and often ends up with significant data transfer overhead leading to degradation of the overall job completion time. In this article, a Resource and Network-aware Data Placement Algorithm (RENDA) is proposed to reduce the non-local executions and thereby reduce the overall job completion time for periodic workloads in the cloud environment. The entire job execution is modeled as a two-stage execution characterized as data distribution and data processing. The RENDA reduces the time of the stages as mentioned above by estimating the heterogeneous performance of the nodes on a real-time basis followed by careful allocation of data in several installments to participating nodes. The experimental results show that the proposed RENDA algorithm consistently outperforms over the recent state-of-the-art alternatives with as much as 28 percent reduction in data transfer overhead leading to 16 percent reduction in average job completion time with 27 percent average speedup on average job execution.

Employing Vertical Elasticity for Efficient Big Data Processing in Container-Based Cloud Environments

Recently, “Big Data” platform technologies have become crucial for distributed processing of diverse unstructured or semi-structured data as the amount of data generated increases rapidly. In order to effectively manage these Big Data, Cloud Computing has been playing an important role by providing scalable data storage and computing resources for competitive and economical Big Data processing. Accordingly, server virtualization technologies that are the cornerstone of Cloud Computing have attracted a lot of research interests. However, conventional hypervisor-based virtualization can cause performance degradation problems due to its heavily loaded guest operating systems and rigid resource allocations. On the other hand, container-based virtualization technology can provide the same level of service faster with a lightweight capacity by effectively eliminating the guest OS layers. In addition, container-based virtualization enables efficient cloud resource management by dynamically adjusting the allocated computing resources (e.g., CPU and memory) during the runtime through “Vertical Elasticity”. In this paper, we present our practice and experience of employing an adaptive resource utilization scheme for Big Data workloads in container-based cloud environments by leveraging the vertical elasticity of Docker, a representative container-based virtualization technique. We perform extensive experiments running several Big Data workloads on representative Big Data platforms: Apache Hadoop and Spark. During the workload executions, our adaptive resource utilization scheme periodically monitors the resource usage patterns of running containers and dynamically adjusts allocated computing resources that could result in substantial improvements in the overall system throughput.

REDUCER: Elimination of Repetitive Codes for Accelerated Iterative Compilation

Low Level Virtual Machine (LLVM) is a widely adopted open source compiler providing numerous optimization opportunities. The discovery of the best optimization sequence in this large space is done via iterative compilation, which incurs substantial overheads, especially for big data applications operating on high volume and variety datasets. The large search space is mostly comprised of identical codes generated via different optimizations. However, no mechanism is implemented inside the LLVM compiler to suppress the redundant testings. In this regard, this paper proposes REDUCER for eliminating the identical code executions by performing Intermediate Representation (IR) level comparisons. REDUCER has been tested using the well-accepted MiCOMP technique in LLVM 3.8 and 9.0 compiler, with embedded (cBench) and big data workloads. In comparison to MiCOMP 19.5 k experiments, REDUCER lowers the experiment count up to 327, i.e. 98 %, and on average to 4 375, i.e. 77 %, for cBench (LLVM-3.8). Similarly, for LLVM-9.0 the reductions are up to 1 931, i.e. 90 %, and on average 5 863, i.e. 69.9 %. Due to the significant experiment reduction, for embedded workloads, the iterative compilation is up to 58.6× and on average 4.1× faster with REDUCER (LLVM-3.8) than MiCOMP, whereas, with REDUCER (LLVM-9.0) the compilation is up to 8.5× and on average 2.9× faster. Moreover, REDUCER is found to be scalable and efficient for big data workloads where the iterative compilation is reduced to few days, as code is compared one time only for a single application tested on multiple datasets.

Green Cloud Software Engineering for Big Data Processing

Internet of Things (IoT) coupled with big data analytics is emerging as the core of smart and sustainable systems which bolsters economic, environmental and social sustainability. Cloud-based data centers provide high performance computing power to analyze voluminous IoT data to provide invaluable insights to support decision making. However, multifarious servers in data centers appear to be the black hole of superfluous energy consumption that contributes to 23% of the global carbon dioxide (CO2) emissions in ICT (Information and Communication Technology) industry. IoT-related energy research focuses on low-power sensors and enhanced machine-to-machine communication performance. To date, cloud-based data centers still face energy–related challenges which are detrimental to the environment. Virtual machine (VM) consolidation is a well-known approach to affect energy-efficient cloud infrastructures. Although several research works demonstrate positive results for VM consolidation in simulated environments, there is a gap for investigations on real, physical cloud infrastructure for big data workloads. This research work addresses the gap of conducting real physical cloud infrastructure-based experiments. The primary goal of setting up a real physical cloud infrastructure is for the evaluation of dynamic VM consolidation approaches which include integrated algorithms from existing relevant research. An open source VM consolidation framework, Openstack NEAT is adopted and experiments are conducted on a Multi-node Openstack Cloud with Apache Spark as the big data platform. Open sourced Openstack has been deployed because it enables rapid innovation, and boosts scalability as well as resource utilization. Additionally, this research work investigates the performance based on service level agreement (SLA) metrics and energy usage of compute hosts. Relevant results concerning the best performing combination of algorithms are presented and discussed.

Bridging Storage Semantics Using Data Labels and Asynchronous I/O

In the era of data-intensive computing, large-scale applications, in both scientific and the BigData communities, demonstrate unique I/O requirements leading to a proliferation of different storage devices and software stacks, many of which have conflicting requirements. Further, new hardware technologies and system designs create a hierarchical composition that may be ideal for computational storage operations. In this article, we investigate how to support a wide variety of conflicting I/O workloads under a single storage system. We introduce the idea of a Label , a new data representation, and, we present LABIOS: a new, distributed, Label- based I/O system. LABIOS boosts I/O performance by up to 17× via asynchronous I/O, supports heterogeneous storage resources, offers storage elasticity, and promotes in situ analytics and software defined storage support via data provisioning. LABIOS demonstrates the effectiveness of storage bridging to support the convergence of HPC and BigData workloads on a single platform.

POLARIS

In this paper, we describe the Polaris distributed SQL query engine in Azure Synapse. It is the result of a multi-year project to re-architect the query processing framework in the SQL DW parallel data warehouse service, and addresses two main goals: (i) converge data warehousing and big data workloads, and (ii) separate compute and state for cloud-native execution. From a customer perspective, these goals translate into many useful features, including the ability to resize live workloads, deliver predictable performance at scale, and to efficiently handle both relational and unstructured data. Achieving these goals required many innovations, including a novel "cell" data abstraction, and flexible, fine-grained, task monitoring and scheduling capable of handling partial query restarts and PB-scale execution. Most importantly, while we develop a completely new scale-out framework, it is fully compatible with T-SQL and leverages decades of investment in the SQL Server single-node runtime and query optimizer. The scalability of the system is highlighted by a 1PB scale run of all 22 TPC-H queries; to our knowledge, this is the first reported run with scale larger than 100TB.

Approximate partition selection for big-data workloads using summary statistics

Many big-data clusters store data in large partitions that support access at a coarse, partition-level granularity. As a result, approximate query processing via row-level sampling is inefficient, often requiring reads of many partitions. In this work, we seek to answer queries quickly and approximately by reading a subset of the data partitions and combining partial answers in a weighted manner without modifying the data layout. We illustrate how to efficiently perform this query processing using a set of pre-computed summary statistics, which inform the choice of partitions and weights. We develop novel means of using the statistics to assess the similarity and importance of partitions. Our experiments on several datasets and data layouts demonstrate that to achieve the same relative error compared to uniform partition sampling, our techniques offer from 2.7x to 70x reduction in the number of partitions read, and the statistics stored per partition require fewer than 100KB.

Real-time resource scaling platform for Big Data workloads on serverless environments

The serverless execution paradigm is becoming an increasingly popular option when workloads are to be deployed in an abstracted way, more specifically, without specifying any infrastructure requirements. Currently, such workloads are typically comprised of small programs or even a series of single functions used as event triggers or to process a data stream. Other applications that may also fit on a serverless scenario are stateless services that may need to seamlessly scale in terms of resources, such as a web server. Although several commercial serverless services are available (e.g., Amazon Lambda), their use cases are mostly limited to the execution of functions or scripts that can be adapted to predefined templates or specifications. However, current research efforts point out that it is interesting for the serverless paradigm to evolve from single functions and support more flexible infrastructure units such as operating-system-level virtualization in the form of containers. In this paper we present a novel platform to automatically scale container resources in real time, while they are running, and without any need for reboots. This platform is evaluated using Big Data workloads, both batch and streaming, as representative examples of applications that could be initially regarded as unsuitable for the serverless paradigm considering the currently available services. The results show how our serverless platform can improve the CPU utilization by up to 77% with an execution time overhead of only 6%, while remaining scalable when using a 32-container cluster.

SARA: Stably and quickly find optimal cloud configurations for heterogeneous big data workloads

With the development of cloud platforms, more and more workloads are operated on the cloud. It is important to find the optimal cloud configurations for these workloads. But most of the previous optimizers concentrate on optimizing single workload. They have high overhead when faced with a large number of workloads to be optimized.To address this problem, we propose SARA, a system that uses Deep Reinforcement Learning to find the optimal cloud configuration for workloads. It is especially stable and quick for any amount of heterogeneous big data workloads.Our research results show that SARA finds the optimal or near-optimal configuration for single workload at a low cost, which is about 28% of Brute Force. For collective workloads, the average probability of SARA to find the optimal configuration is 1.8 times higher than the state-of-the-art collective optimizer Micky, and the cost of SARA for small number of workloads is within 12.5%–80% of Micky.

PerTract: Model Extraction and Specification of Big Data Systems for Performance Prediction by the Example of Apache Spark and Hadoop

Evaluating and predicting the performance of big data applications are required to efficiently size capacities and manage operations. Gaining profound insights into the system architecture, dependencies of components, resource demands, and configurations cause difficulties to engineers. To address these challenges, this paper presents an approach to automatically extract and transform system specifications to predict the performance of applications. It consists of three components. First, a system-and tool-agnostic domain-specific language (DSL) allows the modeling of performance-relevant factors of big data applications, computing resources, and data workload. Second, DSL instances are automatically extracted from monitored measurements of Apache Spark and Apache Hadoop (i.e., YARN and HDFS) systems. Third, these instances are transformed to model- and simulation-based performance evaluation tools to allow predictions. By adapting DSL instances, our approach enables engineers to predict the performance of applications for different scenarios such as changing data input and resources. We evaluate our approach by predicting the performance of linear regression and random forest applications of the HiBench benchmark suite. Simulation results of adjusted DSL instances compared to measurement results show accurate predictions errors below 15% based upon averages for response times and resource utilization.

I can't believe it's not (only) software!

There is a steady increase in the size of data stored and processed as part of data science applications, leading to bottlenecks and inefficiencies at various layers of the stack. One way of reducing such bottlenecks and increasing energy efficiency is by tailoring the underlying distributed storage solution to the application domain, using resources more efficiently. We explore this idea in the context of a popular column-oriented storage format used in big data workloads, namely Apache Parquet. Our prototype uses an FPGA-based storage node that offers high bandwidth data deduplication and a companion software library that exposes an API for Parquet file access. This way the storage node remains general purpose and could be shared by applications from different domains, while, at the same time, benefiting from deduplication well suited to Apache Parquet files and from selective reads of columns in the file. In this demonstration we show, on the one hand, that by relying on the FPGA's dataflow processing model, it is possible to implement in-line deduplication without increasing latencies significantly or reducing throughput. On the other hand, we highlight the benefits of implementing the application-specific aspects in a software library instead of FPGA circuits and how this enables, for instance, regular data science frameworks running in Python to access the data on the storage node and to offload filtering operations.