The new barrier mode in Apache Spark allows embedding distributed deep learning training as a Spark stage to simplify the distributed training workflow. In Spark, a task in a stage does not depend on any other tasks in the same stage, and hence it can be scheduled independently. However, several algorithms require more sophisticated inter-task communications, similar to the MPI paradigm. By combining distributed message passing (using asynchronous network IO), OpenJDK's new auto-vectorization and Spark's barrier execution mode, we can add non-map/reduce based algorithms, such as Cannon's distributed matrix multiplication to Spark. We document an efficient distributed matrix multiplication using Cannon's algorithm, which improves significantly on the performance of the existing MLlib implementation. Used within a barrier task, the algorithm described herein results in an up to 24 percent performance increase on a 10,000x10,000 square matrix with a significantly lower memory footprint. Applications of efficient matrix multiplication include, among others, accelerating the training and implementation of deep convolutional neural network based workloads, and thus such efficient algorithms can play a ground-breaking role in faster, more efficient execution of even the most complicated machine learning tasks.
Execution logs are a crucial medium as they record runtime information of software systems. Although extensive logs are helpful to provide valuable details to identify the root cause in postmortem analysis in case of a failure, this may also incur performance overhead and storage cost. Therefore, in this research, we present the result of our experimental study on seven Spark benchmarks to illustrate the impact of different logging verbosity levels on the execution time and storage cost of distributed software systems. We also evaluate the log effectiveness and the information gain values, and…
Java is the backbone of widely used big data frameworks, such as Apache Spark, due to its productivity, portability from JVM-based execution, and support for a rich set of libraries. However, the performance of these applications can widely vary depending on the runtime flags chosen out of all existing JVM flags. Manually tuning these flags is both cumbersome and error-prone. Automated tuning approaches can ease the task, but current solutions either require considerable processing time or target a subset of flags to avoid time and space requirements. In this paper, we present OneStopTuner, a …
Distributed data processing frameworks (e.g., Hadoop, Spark, and Flink) are widely used to distribute data among computing nodes of a cloud. Recently, there have been increasing efforts aimed at evaluating the performance of distributed data processing frameworks hosted in private and public clouds. However, there is a paucity of research on evaluating the performance of these frameworks hosted in a hybrid cloud, which is an emerging cloud model that integrates private and public clouds to use the best of both worlds. Therefore, in this paper, we evaluate the performance of Hadoop, Spark, and …
Data of the order of terabytes, petabytes, or beyond is known as Big Data. This data cannot be processed using the traditional database software, and hence there comes the need for Big Data Platforms. By combining the capabilities and features of various big data applications and utilities, Big Data Platforms form a single solution. It is a platform that helps to develop, deploy and manage the big data environment. Hadoop and Spark are the two open-source Big Data Platforms provided by Apache. Both these platforms have many configurational parameters, which can have unforeseen effects on the e…
Powerful abstractions such as dataframes are only as efficient as their underlying runtime system. The de-facto distributed data processing framework, Apache Spark, is poorly suited for the modern cloud-based data-science workloads due to its outdated assumptions: static datasets analyzed using coarse-grained transformations. In this paper, we introduce the Indexed DataFrame, an in-memory cache that supports a dataframe abstraction which incorporates indexing capabilities to support fast lookup and join operations. Moreover, it supports appends with multi-version concurrency control. We implem…
We present COPSIM a parallel implementation of standard integer multiplication for the distributed memory setting, and COPK a parallel implementation of Karatsuba's fast integer multiplication algorithm for a distributed memory setting. When using $\mathcal{P}$ processors, each equipped with a local non-shared memory, to compute the product of tho $n$-digits integer numbers, under mild conditions, our algorithms achieve optimal speedup of the computational time. That is, $\mathcal{O}\left(n^2/\mathcal{P}\right)$ for COPSIM, and $\mathcal{O}\left(n^{\log_2 3}/\mathcal{P}\right)$ for COPK. The t…
Distributed dataflow systems like Spark and Flink enable the use of clusters for scalable data analytics. While runtime prediction models can be used to initially select appropriate cluster resources given target runtimes, the actual runtime performance of dataflow jobs depends on several factors and varies over time. Yet, in many situations, dynamic scaling can be used to meet formulated runtime targets despite significant performance variance. This paper presents Enel, a novel dynamic scaling approach that uses message propagation on an attributed graph to model dataflow jobs and, thus, allo…
The A64FX CPU powers the current number one supercomputer on the Top500 list. Although it is a traditional cache-based multicore processor, its peak performance and memory bandwidth rival accelerator devices. Generating efficient code for such a new architecture requires a good understanding of its performance features. Using these features, we construct the Execution-Cache-Memory (ECM) performance model for the A64FX processor in the FX700 supercomputer and validate it using streaming loops. We also identify architectural peculiarities and derive optimization hints. Applying the ECM model to …
FalconGEMM: Surpassing Hardware Peaks with Lower-Complexity Matrix Multiplication
Tencent, Shenzhen, China（腾讯）；The Hong Kong Polytechnic University, Hong Kong（香港理工大学）；NVIDIA, Beijing, China（英伟达）；Shanghai Innovation Institute, Shanghai, China（上海创新研究院）；RIKEN, Tokyo, Japan（理化学研究所）；Southern University of Science and Technology, Shenzhen, China（南方科技大学）；Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China（中国科学院深圳先进技术研究院）
Sparse general matrix multiplication (SpGEMM) is an important and expensive computation primitive in many real-world applications. Due to SpGEMM's inherent irregularity and the vast diversity of its input matrices, developing high-performance SpGEMM implementation on modern processors such as GPUs is challenging. The state-of-the-art SpGEMM libraries (i.e., $nsparse$ and $spECK$) adopt several algorithms to tackle the challenges of global load balance, local load balance, and allocation of the result matrix. While these libraries focus on the high-level algorithm design for SpGEMM, they neglec…
PingAn: An Insurance Scheme for Job Acceleration in Geo-distributed Big Data Analytics System
云边缘系统中的地理分布数据分析正在成为日常需求。为了节省广域数据传输的时间，一些任务被分散到边缘。然而，由于计算能力有限、负载干扰和簇级不可达等问题，边缘作业难以有效执行，影响了作业的效率和可靠性。跨集群启动副本可以确保任务的完成。考虑到集群异构性和伴随的远程数据获取，副本的集群选择会影响执行质量，因为不同的保险计划会带来不同的收入。为了提供在线实时分析结果，系统需要确保到达作业的地理分布资源。我们的挑战是通过动态权衡保险收益与占用额外保险资源的损失来实现最优收益。为此，我们设计了平安，一个在线保险算法承诺$(1+\varepsilon)\!\!-\!speed \ o(\frac{1}{\varepsilon^2+\varepsilon})\!\!-\!competitive$的工作流时间的总和，通过跨集群复制的任务。PingAn在具有最小未处理数据量的作业的前部分中共享资源，并且该部分是可调节的以适应系统负载条件。共享后，平安对任务进行具体保障，遵循效率优先、可靠性感知的原则，优化副本收益对任务性能的影响。跟踪驱动的模拟表明，平安可以减少平均作业流时间至少$14\%$比国家的最先进的投机机制。并在Spark on Yarn系统中实现了平安，验证了该系统的实用性和通用性。实验表明，与默认的Spark执行相比，PingAn可以将平均作业流程时间减少多达$40\%$。
In the big data era, the key feature that each algorithm needs to have is the possibility of efficiently running in parallel in a distributed environment. The popular Silhouette metric to evaluate the quality of a clustering, unfortunately, does not have this property and has a quadratic computational complexity with respect to the size of the input dataset. For this reason, its execution has been hindered in big data scenarios, where clustering had to be evaluated otherwise. To fill this gap, in this paper we introduce the first algorithm that computes the Silhouette metric with linear comple…
A novel fault-tolerant computation technique based on array Belief Propagation (BP)-decodable XOR (BP-XOR) codes is proposed for distributed matrix-matrix multiplication. The proposed scheme is shown to be configurable and suited for modern hierarchical compute architectures such as Graphical Processing Units (GPUs) equipped with multiple nodes, whereby each has many small independent processing units with increased core-to-core communications. The proposed scheme is shown to outperform a few of the well--known earlier strategies in terms of total end-to-end execution time while in presence of…
Sparse matrix-vector multiplication is often employed in many data-analytic workloads in which low latency and high throughput are more valuable than exact numerical convergence. FPGAs provide quick execution times while offering precise control over the accuracy of the results thanks to reduced-precision fixed-point arithmetic. In this work, we propose a novel streaming implementation of Coordinate Format (COO) sparse matrix-vector multiplication, and study its effectiveness when applied to the Personalized PageRank algorithm, a common building block of recommender systems in e-commerce websi…
Several manufacturers have already started to commercialize near-bank Processing-In-Memory (PIM) architectures. Near-bank PIM architectures place simple cores close to DRAM banks and can yield significant performance and energy improvements in parallel applications by alleviating data access costs. Real PIM systems can provide high levels of parallelism, large aggregate memory bandwidth and low memory access latency, thereby being a good fit to accelerate the widely-used, memory-bound Sparse Matrix Vector Multiplication (SpMV) kernel. This paper provides the first comprehensive analysis of SpM…
Sparse general matrix multiplication (SpGEMM) is a fundamental building block in numerous scientific applications. One critical task of SpGEMM is to compute or predict the structure of the output matrix (i.e., the number of nonzero elements per output row) for efficient memory allocation and load balance, which impact the overall performance of SpGEMM. Existing work either precisely calculates the output structure or adopts upper-bound or sampling-based methods to predict the output structure. However, these methods either take much execution time or are not accurate enough. In this paper, we …
Several manufacturers have already started to commercialize near-bank Processing-In-Memory (PIM) architectures. Near-bank PIM architectures place simple cores close to DRAM banks and can yield significant performance and energy improvements in parallel applications by alleviating data access costs. Real PIM systems can provide high levels of parallelism, large aggregate memory bandwidth and low memory access latency, thereby being a good fit to accelerate the widely-used, memory-bound Sparse Matrix Vector Multiplication (SpMV) kernel. This paper provides the first comprehensive analysis of SpM…
