DOCSLIB.ORG

Portable Stateful Big Data Processing in Apache Beam

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Portable Stateful Big Data Processing in Apache Beam Portable stateful big data processing in Apache Beam Kenneth Knowles Apache Beam PMC Software Engineer @ Google https://s.apache.org/ffsf-2017-beam-state [email protected] / @KennKnowles Flink Forward San Francisco 2017 Agenda 1. What is Apache Beam? 2. State 3. Timers 4. Example & Little Demo What is Apache Beam? TL;DR (Flink draws it more like this) 4 DAGs, DAGs, DAGs Apache Beam Apache Flink Apache Cloud Hadoop Apache Apache Dataflow Spark Samza MapReduce Apache Apache Apache (paper) Storm Gearpump Apex (incubating) FlumeJava (paper) Heron MillWheel (paper) Dataflow Model (paper) 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Apache Flink local, on-prem, The Beam Vision cloud Cloud Dataflow: Java fully managed input.apply( Apache Spark Sum.integersPerKey()) local, on-prem, cloud Sum Per Key Apache Apex Python local, on-prem, cloud input | Sum.PerKey() Apache Gearpump (incubating) ⋮ ⋮ 6 Apache Flink local, on-prem, The Beam Vision cloud Cloud Dataflow: Python fully managed input | KakaIO.read() Apache Spark local, on-prem, cloud KafkaIO Apache Apex ⋮ local, on-prem, cloud Apache Java Gearpump (incubating) class KafkaIO extends UnboundedSource { … } ⋮ 7 The Beam Model PTransform Pipeline PCollection (bounded or unbounded) 8 The Beam Model What are you computing? (read, map, reduce) Where in event time? (event time windowing) When in processing time are results produced? (triggers) How do refinements relate? (accumulation mode) 9 What are you computing? Read ParDo Grouping Composite Parallel connectors to Per element Group by key, Encapsulated external systems "Map" Combine per key, subgraph "Reduce" 10 State What are you computing? Read ParDo Grouping Composite Parallel connectors to Per element Group by key, Encapsulated external systems "Map" Combine per key, subgraph "Reduce" 12 State for ParDo ParDo(DoFn) 13 ParDo.of(new DoFn<...>() { "Example" // declare some state @ProcessElement public void process(...) { // update quantiles // output if needed } }) Per Key Quantiles 14 Partitioned Quantiles Quantiles Quantiles 15 Parallelism! new DoFn<Foo, Quantiles<Foo>>() { @StateId("quantiles") private final StateSpec<...> quantilesState = StateSpecs.combining(); … } Quantiles Quantiles Quantiles DoFn DoFn DoFn 16 Windowed Window into Fixed windows of one hour Quantiles Quantiles Quantiles DoFn DoFn DoFn Expected result: Quantiles for each hour 17 Windowed Window into windows of 30 min sliding by 10 min Quantiles Quantiles Quantiles DoFn DoFn DoFn Expected result: Quantiles for 30 minutes sliding by 10 min 18 State is per key and window ... <k, w>1 <k, w>2 <k, w>3 "x" 3 7 15 "y" "fizz" "7" "fizzbuzz" ... Bonus: automatically garbage collected when a window expires (vs manual clearing of keyed state) Windowed Window into Fixed windows of one hour Quantiles Quantiles Quantiles DoFn DoFn DoFn Expected result: Quantiles for each hour 20 What about Combine? Window into Fixed windows of one hour Quantiles Quantiles Quantiles Combiner Combiner Combiner Expected result: Quantiles for each hour 21 Combine vs State (naive conceptualization) Window hourly Window hourly Quantiles Quantiles Combiner DoFn Expected result: Expected result: Quantiles for each hour Quantiles for each hour 22 Combine vs State (likely execution plan) Quantiles Quantiles non-associative* Combiner Combiner non-commutative* Shuffle Elements multi-output Shuffle side outputs Accumulators (user is in control) Quantiles Quantiles associative Combiner DoFn commutative single-output enables optimizations (engine is in control) Combine vs State Quantiles Combiner Quantiles DoFn output governed by trigger (data/computation unaware) "output only when there's an interesting change" (data/computation aware) Example: Arbitrary indices E B A C D non-associative Assign indices non-commutative non-deterministic E,4 B,3 and totally fine! A,2 C,1 D,0 Kinds of state ● Value - just a mutable cell for a value ● Bag - supports "blind writes" ● Combining - has a CombineFn built in; can support blind writes and lazy accumulation ● Set - membership checking ● Map - lookups and partial writes Timers new DoFn<...>() { Timers for ParDo @TimerId("timeout") private final TimerSpec timeoutTimer = TimerSpecs.timer(TimeDomain.PROCESSING_TIME); @OnTimer("timout") public void timeout(...) { // access state, set new timers, output } … } Stateful ParDo 28 Timers in processing time buffer request "call me back in 5 On Element seconds" On Timer batched RPC output based only output return on incoming value of batched element RPC Timers in event time "call me back when store speculative result the watermark hits the end of the On Element window" On Timer output output replacement speculative result value only if immediately changed More example uses for state & timers ● Per-key arbitrary numbering ● Output only when result changes ● Tighter "side input" management for slowly changing dimension ● Streaming join-matrix / join-biclique ● Fine-grained combine aggregation and output control ● Per-key "workflows" like user sign up flow w/ expiration ● Low-latency deduplication (let the first through, squash the rest) Performance considerations (cross-runner) ● Shuffle to colocate keys ● Linear processing of elements for key+window ● Window merging ● Storage of state and timers ● GC of state 32 Demo 33 Summary State and Timers in Beam... ● … unlock new uses cases ● … they "just work" with event time windowing ● … are portable across runners (implementation ongoing) Thank you for listening! This talk: ● Me - @KennKnowles ● These Slides - https://s.apache.org/ffsf-2017-beam-state Go Deeper ● Design doc - https://s.apache.org/beam-state ● Blog post - https://beam.apache.org/blog/2017/02/13/stateful-processing.html Join the Beam community: ● User discussions - [email protected] ● Development discussions - [email protected] ● Follow @ApacheBeam on Twitter You can contribute to Beam + Flink ● New types of state ● Easy launch of Beam job on Flink-on-YARN ● Integration tests at scale ● Fit and finish: polish, polish, polish! ● … and lots more! https://beam.apache.org.
Load more

Recommended publications
  • Apache Flink™: Stream and Batch Processing in a Single Engine
    Apache Flink™: Stream and Batch Processing in a Single Engine Paris Carboney Stephan Ewenz Seif Haridiy Asterios Katsifodimos* Volker Markl* Kostas Tzoumasz yKTH & SICS Sweden zdata Artisans *TU Berlin & DFKI parisc,[email protected] fi[email protected] fi[email protected] Abstract Apache Flink1 is an open-source system for processing streaming and batch data. Flink is built on the philosophy that many classes of data processing applications, including real-time analytics, continu- ous data pipelines, historic data processing (batch), and iterative algorithms (machine learning, graph analysis) can be expressed and executed as pipelined fault-tolerant dataflows. In this paper, we present Flink’s architecture and expand on how a (seemingly diverse) set of use cases can be unified under a single execution model. 1 Introduction Data-stream processing (e.g., as exemplified by complex event processing systems) and static (batch) data pro- cessing (e.g., as exemplified by MPP databases and Hadoop) were traditionally considered as two very different types of applications. They were programmed using different programming models and APIs, and were exe- cuted by different systems (e.g., dedicated streaming systems such as Apache Storm, IBM Infosphere Streams, Microsoft StreamInsight, or Streambase versus relational databases or execution engines for Hadoop, including Apache Spark and Apache Drill). Traditionally, batch data analysis made up for the lion’s share of the use cases, data sizes, and market, while streaming data analysis mostly served specialized applications. It is becoming more and more apparent, however, that a huge number of today’s large-scale data processing use cases handle data that is, in reality, produced continuously over time.
    [Show full text]
  • Evaluation of SPARQL Queries on Apache Flink
    applied sciences Article SPARQL2Flink: Evaluation of SPARQL Queries on Apache Flink Oscar Ceballos 1 , Carlos Alberto Ramírez Restrepo 2 , María Constanza Pabón 2 , Andres M. Castillo 1,* and Oscar Corcho 3 1 Escuela de Ingeniería de Sistemas y Computación, Universidad del Valle, Ciudad Universitaria Meléndez Calle 13 No. 100-00, Cali 760032, Colombia; [email protected] 2 Departamento de Electrónica y Ciencias de la Computación, Pontificia Universidad Javeriana Cali, Calle 18 No. 118-250, Cali 760031, Colombia; [email protected] (C.A.R.R.); [email protected] (M.C.P.) 3 Ontology Engineering Group, Universidad Politécnica de Madrid, Campus de Montegancedo, Boadilla del Monte, 28660 Madrid, Spain; ocorcho@fi.upm.es * Correspondence: [email protected] Abstract: Existing SPARQL query engines and triple stores are continuously improved to handle more massive datasets. Several approaches have been developed in this context proposing the storage and querying of RDF data in a distributed fashion, mainly using the MapReduce Programming Model and Hadoop-based ecosystems. New trends in Big Data technologies have also emerged (e.g., Apache Spark, Apache Flink); they use distributed in-memory processing and promise to deliver higher data processing performance. In this paper, we present a formal interpretation of some PACT transformations implemented in the Apache Flink DataSet API. We use this formalization to provide a mapping to translate a SPARQL query to a Flink program. The mapping was implemented in a prototype used to determine the correctness and performance of the solution. The source code of the Citation: Ceballos, O.; Ramírez project is available in Github under the MIT license.
    [Show full text]
  • Regeldokument
    Master’s degree project Source code quality in connection to self-admitted technical debt Author: Alina Hrynko Supervisor: Morgan Ericsson Semester: VT20 Subject: Computer Science Abstract The importance of software code quality is increasing rapidly. With more code being written every day, its maintenance and support are becoming harder and more expensive. New automatic code review tools are developed to reach quality goals. One of these tools is SonarQube. However, people keep their leading role in the development process. Sometimes they sacrifice quality in order to speed up the development. This is called Technical Debt. In some particular cases, this process can be admitted by the developer. This is called Self-Admitted Technical Debt (SATD). Code quality can also be measured by such static code analysis tools as SonarQube. On this occasion, different issues can be detected. The purpose of this study is to find a connection between code quality issues, found by SonarQube and those marked as SATD. The research questions include: 1) Is there a connection between the size of the project and the SATD percentage? 2) Which types of issues are the most widespread in the code, marked by SATD? 3) Did the introduction of SATD influence the bug fixing time? As a result of research, a certain percentage of SATD was found. It is between 0%–20.83%. No connection between the size of the project and the percentage of SATD was found. There are certain issues that seem to relate to the SATD, such as “Duplicated code”, “Unused method parameters should be removed”, “Cognitive Complexity of methods should not be too high”, etc.
    [Show full text]
  • Trifacta Data Preparation for Amazon Redshift and S3 Must Be Deployed Into an Existing Virtual Private Cloud (VPC)
    Install Guide for Data Preparation for Amazon Redshift and S3 Version: 7.1 Doc Build Date: 05/26/2020 Copyright © Trifacta Inc. 2020 - All Rights Reserved. CONFIDENTIAL These materials (the “Documentation”) are the confidential and proprietary information of Trifacta Inc. and may not be reproduced, modified, or distributed without the prior written permission of Trifacta Inc. EXCEPT AS OTHERWISE PROVIDED IN AN EXPRESS WRITTEN AGREEMENT, TRIFACTA INC. PROVIDES THIS DOCUMENTATION AS-IS AND WITHOUT WARRANTY AND TRIFACTA INC. DISCLAIMS ALL EXPRESS AND IMPLIED WARRANTIES TO THE EXTENT PERMITTED, INCLUDING WITHOUT LIMITATION THE IMPLIED WARRANTIES OF MERCHANTABILITY, NON-INFRINGEMENT AND FITNESS FOR A PARTICULAR PURPOSE AND UNDER NO CIRCUMSTANCES WILL TRIFACTA INC. BE LIABLE FOR ANY AMOUNT GREATER THAN ONE HUNDRED DOLLARS ($100) BASED ON ANY USE OF THE DOCUMENTATION. For third-party license information, please select About Trifacta from the Help menu. 1. Quick Start . 4 1.1 Install from AWS Marketplace . 4 1.2 Upgrade for AWS Marketplace . 7 2. Configure . 8 2.1 Configure for AWS . 8 2.1.1 Configure for EC2 Role-Based Authentication . 14 2.1.2 Enable S3 Access . 16 2.1.2.1 Create Redshift Connections 28 3. Contact Support . 30 4. Legal 31 4.1 Third-Party License Information . 31 Page #3 Quick Start Install from AWS Marketplace Contents: Product Limitations Internet access Install Desktop Requirements Pre-requisites Install Steps - CloudFormation template SSH Access Troubleshooting SELinux Upgrade Documentation Related Topics This guide steps through the requirements and process for installing Trifacta® Data Preparation for Amazon Redshift and S3 through the AWS Marketplace.
    [Show full text]
  • Apache Flink Fast and Reliable Large-Scale Data Processing
    Apache Flink Fast and Reliable Large-Scale Data Processing Fabian Hueske @fhueske 1 What is Apache Flink? Distributed Data Flow Processing System • Focused on large-scale data analytics • Real-time stream and batch processing • Easy and powerful APIs (Java / Scala) • Robust execution backend 2 What is Flink good at? It‘s a general-purpose data analytics system • Real-time stream processing with flexible windows • Complex and heavy ETL jobs • Analyzing huge graphs • Machine-learning on large data sets • ... 3 Flink in the Hadoop Ecosystem Libraries Dataflow Table API Table ML Library ML Gelly Library Gelly Apache MRQL Apache SAMOA DataSet API (Java/Scala) DataStream API (Java/Scala) Flink Core Optimizer Stream Builder Runtime Environments Embedded Local Cluster Yarn Apache Tez Data HDFS Hadoop IO Apache HBase Apache Kafka Apache Flume Sources HCatalog JDBC S3 RabbitMQ ... 4 Flink in the ASF • Flink entered the ASF about one year ago – 04/2014: Incubation – 12/2014: Graduation • Strongly growing community 120 100 80 60 40 20 0 Nov.10 Apr.12 Aug.13 Dec.14 #unique git committers (w/o manual de-dup) 5 Where is Flink moving? A "use-case complete" framework to unify batch & stream processing Data Streams • Kafka Analytical Workloads • RabbitMQ • ETL • ... • Relational processing Flink • Graph analysis • Machine learning “Historic” data • Streaming data analysis • HDFS • JDBC • ... Goal: Treat batch as finite stream 6 Programming Model & APIs HOW TO USE FLINK? 7 Unified Java & Scala APIs • Fluent and mirrored APIs in Java and Scala • Table API
    [Show full text]
  • The Forrester Wave™: Streaming Analytics, Q3 2019 the 11 Providers That Matter Most and How They Stack up by Mike Gualtieri September 23, 2019
    LICENSED FOR INDIVIDUAL USE ONLY The Forrester Wave™: Streaming Analytics, Q3 2019 The 11 Providers That Matter Most And How They Stack Up by Mike Gualtieri September 23, 2019 Why Read This Report Key Takeaways In our 26-criterion evaluation of streaming Software AG, IBM, Microsoft, Google, And analytics providers, we identified the 11 most TIBCO Software Lead The Pack significant ones — Alibaba, Amazon Web Forrester’s research uncovered a market in which Services, Cloudera, EsperTech, Google, IBM, Software AG, IBM, Microsoft, Google, and TIBCO Impetus, Microsoft, SAS, Software AG, and Software are Leaders; Cloudera, SAS, Amazon TIBCO Software — and researched, analyzed, Web Services, and Impetus are Strong Performers; and scored them. This report shows how each and EsperTech and Alibaba are Contenders. provider measures up and helps application Analytics Prowess, Scalability, And development and delivery (AD&D) professionals Deployment Freedom Are Key Differentiators select the right one for their needs. Depth and breadth of analytics types on streaming data are critical. But that is all for naught if streaming analytics vendors cannot also scale to handle potentially huge volumes of streaming data. Also, it’s critical that streaming analytics can be deployed where it is most needed, such as on-premises, in the cloud, and/ or at the edge. This PDF is only licensed for individual use when downloaded from forrester.com or reprints.forrester.com. All other distribution prohibited. FORRESTER.COM FOR APPLICATION DEVELOPMENT & DELIVERY PROFESSIONALS The Forrester Wave™: Streaming Analytics, Q3 2019 The 11 Providers That Matter Most And How They Stack Up by Mike Gualtieri with Srividya Sridharan and Robert Perdoni September 23, 2019 Table Of Contents Related Research Documents 2 Enterprises Must Take A Streaming-First The Future Of Machine Learning Is Unstoppable Approach To Analytics Predictions 2019: Artificial Intelligence 3 Evaluation Summary Predictions 2019: Business Insights 6 Vendor Offerings 6 Vendor Profiles Leaders Share reports with colleagues.
    [Show full text]
  • The Ubiquity of Large Graphs and Surprising Challenges of Graph Processing: Extended Survey
    The VLDB Journal https://doi.org/10.1007/s00778-019-00548-x SPECIAL ISSUE PAPER The ubiquity of large graphs and surprising challenges of graph processing: extended survey Siddhartha Sahu1 · Amine Mhedhbi1 · Semih Salihoglu1 · Jimmy Lin1 · M. Tamer Özsu1 Received: 21 January 2019 / Revised: 9 May 2019 / Accepted: 13 June 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019 Abstract Graph processing is becoming increasingly prevalent across many application domains. In spite of this prevalence, there is little research about how graphs are actually used in practice. We performed an extensive study that consisted of an online survey of 89 users, a review of the mailing lists, source repositories, and white papers of a large suite of graph software products, and in-person interviews with 6 users and 2 developers of these products. Our online survey aimed at understanding: (i) the types of graphs users have; (ii) the graph computations users run; (iii) the types of graph software users use; and (iv) the major challenges users face when processing their graphs. We describe the participants’ responses to our questions highlighting common patterns and challenges. Based on our interviews and survey of the rest of our sources, we were able to answer some new questions that were raised by participants’ responses to our online survey and understand the specific applications that use graph data and software. Our study revealed surprising facts about graph processing in practice. In particular, real-world graphs represent a very diverse range of entities and are often very large, scalability and visualization are undeniably the most pressing challenges faced by participants, and data integration, recommendations, and fraud detection are very popular applications supported by existing graph software.
    [Show full text]
  • Scalable and Flexible Middleware for Dynamic Data Flows
    SCALABLEANDFLEXIBLEMIDDLEWAREFORDYNAMIC DATAFLOWS stephan boomker Master thesis Computing Science Software Engineering and Distributed Systems Primaray RuG supervisor : Prof. Dr. M. Aiello Secondary RuG supervisor : Prof. Dr. P. Avgeriou Primary TNO supervisor : MSc. E. Harmsma Secondary TNO supervisor : MSc. E. Lazovik August 16, 2016 – version 1.0 [ August 16, 2016 at 11:47 – classicthesis version 1.0 ] Stephan Boomker: Scalable and flexible middleware for dynamic data flows, Master thesis Computing Science, © August 16, 2016 [ August 16, 2016 at 11:47 – classicthesis version 1.0 ] ABSTRACT Due to the concepts of Internet of Things and Big data, the traditional client-server architecture is not sufficient any more. One of the main reasons is wide range of expanding heterogeneous applications, data sources and environments. New forms of data processing require new architectures and techniques in order to be scalable, flexible and able to handle fast dynamic data flows. The backbone of all those objects, applications and users is called the middleware. This research goes about designing and implementing a middle- ware by taking into account different state of the art tools and tech- niques. To come up to a solution which is able to handle a flexible set of sources and models across organizational borders. At the same time it is de-centralized, distributed and, although de-central able to perform semantic based system integration centrally. This is accom- plished by introducing of an architecture containing a combination of data integration patterns, semantic storage and stream processing patterns. A reference implementation is presented of the proposed architec- ture based on Apache Camel framework. This prototype provides the ability to dynamically create and change flexible and distributed data flows during runtime.
    [Show full text]
  • Big Data Analysis Using Hadoop Lecture 4 Hadoop Ecosystem
    1/16/18 Big Data Analysis using Hadoop Lecture 4 Hadoop EcoSystem Hadoop Ecosytems 1 1/16/18 Overview • Hive • HBase • Sqoop • Pig • Mahoot / Spark / Flink / Storm • Hadoop & Data Management Architectures Hive 2 1/16/18 Hive • Data Warehousing Solution built on top of Hadoop • Provides SQL-like query language named HiveQL • Minimal learning curve for people with SQL expertise • Data analysts are target audience • Ability to bring structure to various data formats • Simple interface for ad hoc querying, analyzing and summarizing large amountsof data • Access to files on various data stores such as HDFS, etc Website : http://hive.apache.org/ Download : http://hive.apache.org/downloads.html Documentation : https://cwiki.apache.org/confluence/display/Hive/LanguageManual Hive • Hive does NOT provide low latency or real-time queries • Even querying small amounts of data may take minutes • Designed for scalability and ease-of-use rather than low latency responses • Translates HiveQL statements into a set of MapReduce Jobs which are then executed on a Hadoop Cluster • This is changing 3 1/16/18 Hive • Hive Concepts • Re-used from Relational Databases • Database: Set of Tables, used for name conflicts resolution • Table: Set of Rows that have the same schema (same columns) • Row: A single record; a set of columns • Column: provides value and type for a single value • Can can be dived up based on • Partitions • Buckets Hive – Let’s work through a simple example 1. Create a Table 2. Load Data into a Table 3. Query Data 4. Drop a Table 4 1/16/18
    [Show full text]
  • Researching Algorithmic Institutions Essay
    Researching Algorithmic Institutions RESEARCHING ALGORITHMIC governance produces an essentially harnessed to the pursuit of procedures experimental condition of institutionality. That and realization of rules. Experiments lend INSTITUTIONS which becomes available for ‘disruption’ or themselves to the goal-oriented world of Liam Magee and Ned Rossiter ‘innovation’ — both institutional encodings — algorithms. As such, the invention of new Like any research centre that today is equally prescribed within silicon test-beds institutional forms and practices would seem 3 investigates the media conditions of social that propose limits to political possibility. antithetical to experiments in algorithmic organization, the Institute for Culture and Experiments privilege the repeatability and governance. Yet what if we consider Society modulates functional institutional reproducibility of action. This is characteristic experience itself as conditioned and governance with what might be said to be, of algorithmic routines that accommodate made possible by experiments in operatively and with a certain conscientious variation only through the a priori of known algorithmic governance? attention to method, algorithmic experiments. statistical parameters. Innovation, in other Surely enough, the past few decades have In this essay we convolve these terms. words, is merely a variation of the known seen a steady transformation of many As governance moves beyond Weberian within the horizon of fault tolerance. institutional settings. There are many studies proceduralism toward its algorithmic Experiments in algorithmic governance that account for such change as coinciding automation, research life itself becomes are radically dissimilar from the experience with and often directly resulting from the ways subject to institutional experimentation. of politics and culture, which can be in which neoliberal agendas have variously Parametric adjustment generates sine wave- understood as the constitutive outside of impacted organizational values and practices.
    [Show full text]
  • CIF21 Dibbs: Middleware and High Performance Analytics Libraries for Scalable Data Science
    NSF14-43054 start October 1, 2014 Datanet: CIF21 DIBBs: Middleware and High Performance Analytics Libraries for Scalable Data Science • Indiana University (Fox, Qiu, Crandall, von Laszewski), • Rutgers (Jha) • Virginia Tech (Marathe) • Kansas (Paden) • Stony Brook (Wang) • Arizona State(Beckstein) • Utah(Cheatham) Overview by Gregor von Laszewski April 6 2016 http://spidal.org/ http://news.indiana.edu/releases/iu/2014/10/big-data-dibbs-grant.shtml http://www.nsf.gov/awardsearch/showAward?AWD_ID=1443054 04/6/2016 1 Some Important Components of SPIDAL Dibbs • NIST Big Data Application Analysis: features of data intensive apps • HPC-ABDS: Cloud-HPC interoperable software with performance of HPC (High Performance Computing) and the rich functionality of the commodity Apache Big Data Stack. – Reservoir of software subsystems – nearly all from outside project and mix of HPC and Big Data communities – Leads to Big Data – Simulation - HPC Convergence • MIDAS: Integrating Middleware – from project • Applications: Biomolecular Simulations, Network and Computational Social Science, Epidemiology, Computer Vision, Spatial Geographical Information Systems, Remote Sensing for Polar Science and Pathology Informatics. • SPIDAL (Scalable Parallel Interoperable Data Analytics Library): Scalable Analytics for – Domain specific data analytics libraries – mainly from project – Add Core Machine learning Libraries – mainly from community – Performance of Java and MIDAS Inter- and Intra-node • Benchmarks – project adds to community; See WBDB 2015 Seventh Workshop
    [Show full text]
  • Issues at the Intersection of AI, Streaming, HPC, Data Centers And
    Issues at the intersection of AI, Streaming, HPC, Data-centers and the Edge NITRD Middleware and Grid Interagency Coordination Team (MAGIC) Geoffrey Fox 1 April 2020 Digital Science Center, Indiana University [email protected], http://www.dsc.soic.indiana.edu/ Digital Science Center AI, Streaming, HPC 1 Some Simple Observations • Consider Science Research Benchmarks in MLPerf • Enhance collaboration between Industry and Research; HPC and MLPerf MLSys communities • Support common environments from Edge to Cloud and HPC systems • Huge switch to Deep Learning for Big Data • Many new algorithms to be developed • Deep Learning for (Geospatial) Time Series (staple of the edge) incredibly promising: obvious relevance to Covid-19 studies • Examples • Inference at the edge • Fusion instabilities • Ride-hailing • Racing Cars • Images • Earthquakes • Solving ODE’s • Particle Physics Events • Timely versus real-time (throughput versus latency); both important Digital Science Center AI, Streaming, HPC 2 Sam,bo!"~va ....... H SIGOPT skymiz•e ®·,,.,· - SiMa"1 MLPerf Consortium Deep Learning Benchmarks Rpa2a1 Sambanova Samsung S.LSI Sigopt SiMaAI Skymizer Supermicro Tencent synOPSYS" TTE.NSYR TERAa,'NE Q>r.. .._, ... ..,. ~ VHind Companies •m verifAI Some Relevant Working Groups Synopsys Tencent Tensyr Teradyne Transpire Ventures VerifAI VMind , Al labs.tw •• 11--.u••--· a-- arm • Training vmware ,( volley.com VV/\V:. £ XILINX. Al Labs.tw Alibaba AMO Andes Technology Aon Devices Arm Baidu • Inference (Batch and Streaming) VMware Volley Wave Computing Wiwynn WekalO Xilinx • TinyML (embedded) !.!!P.l!!'.!!!!!! cadence' CALYPSO {in]i!ur f erebras CEVA • Deep Learning for Time Series Resea rchers from BAAi Cadence Calypso Al Centaur Technology Cerebras Ceva Cirrus • Power ... dii .. 1,.
    [Show full text]