API Whitepaper Selection

Total Page:16

File Type:pdf, Size:1020Kb

API Whitepaper Selection The API revolution continues! API Whitepaper Selection 25+ pages of knowledge for API enthusiasts. Learn about exciting topics such as API Development, Management, Design and Platforms & Software as a Service API Conference @api_conference @apiconf #APICON apiconference.net content API Design The demand for realtime APIs 3 Realtime API Family helps us to reduce complexity by Kieran Kilbride-Singh API Development What APIs Can and Cannot Do 6 How to avoid common API misconceptions and the resulting disillusionment by Erik Wilde Go for gRPC? 11 Develop gRPC services with Go by Rainer Stropek API Platforms & Software-as-a-Service X-Road – The Free and Open Source Data Exchange Layer 17 A centrally managed distributed Data Exchange Layer between information systems by Petteri Kivimäki Regulation as a Service 21 Rethinking APIs by Khallai Taylor API Management Shopware 6 and the new API-First Approach 23 Putting the API on the Acid Test by Thomas Eiling APIs have Properties – find and test them! 26 Taking executable specifications seriously by Dr. Michael Sperber WHITEPAPER API Design Realtime API Family helps us to reduce complexity The demand for realtime APIs Looking to 2020 and beyond, the proportion of data produced and consumed in realtime is growing exponentially by Kieran Kilbride-Singh defines it as anything that happens in under 100ms as this is typically the threshold for perceiving something IDC predict that by 2025 1/3 of all data produced glob- as happening immediately. ally will be realtime. The engineers and organizations Any API designed such that data flowing from pro- that make up the realtime ecosystem, Ably included, ducers to consumers happens in the shortest amount of have yet to agree on how we describe the APIs we’re time possible can therefore be described as realtime. creating and consuming that are powering this growth. Ably has settled on realtime API as an umbrella for The problem is that there are various ways to describe event-driven, streaming, pub/sub, push, and other APIs. APIs that provide realtime functionality. We’ve seen many more than four ways of describing API evangelist Kin Lane, and many others, have been this, but these are the most popular. It’s why we call it writing and talking about these new types of APIs for a the Realtime API Family (Image 1). while. Event-driven seems to be the most common descriptor. Gartner has adopted the term, stating that by 2020 50 % of all managed APIs will be event-driven. But still today there is no consistency or consensus between what terms like realtime API, event-driven API, or streaming API refer to. Often they’re used interchangeably. As an engineering team and API provid- er working on a global pub/sub messaging platform, we work with these ideas every- day. Talking to our users we see the lan- guage they use to describe both what we do and what they achieve using our APIs. And we’ve watched discussions around event-driven architecture, webhooks, and streaming data proliferate. Over the years we’ve thought extensively about the best terminology to use and arrived at what we call the Realtime API Family Defining realtime When it comes to realtime in the context of user interaction, it can be defined as a function or interaction perceived as imme- diate. As a concept in this context then, re- altime is easy enough to understand. Ably Image 1: Realtime API Family www.apiconference.net 3 WHITEPAPER API Design Image 2: Ably’s Pub/Sub Channels API acts as a serverless message bus Streaming APIs Push APIs A stream is a means of transporting data. Streaming is a Push is another umbrella term for every API that is real consumer pattern that describes how consumers receive time. events through a stream. But it’s just a producer messaging pattern. It means A streaming API will commonly address issues of data that data is pushed upstream over a connection, versus integrity with: a pull mechanism used by the request/ response pat- tern. • Message ordering - ensure messages are delivered in An example is helpful. A push notifications API (Im- the order they were published age 3) is an example of a push API. Or a sporting event • Stream continuity / resume - upon disconnection, might send a single score update that traverses multiple resume from the disconnected point within a set pe- systems via push APIs, resulting in millions of messages riod of time to global fans within a few hundred milliseconds. • Contiguous serial numbers - a simple series of `ACK` Push can also refer to triggering a new request as with a or `NACK` responses, each addressing a contiguous Webhook. Or it can mean ‘push subscription’ – i.e. a pro- sequence of messages ducer needs to reach out to a consumer. WebSub is an ex- ample of this. But that’s beyond the scope of this article. Kafka offers a streaming API following this model for internal systems. Event-driven APIs Event-driven is an architectural design pattern that de- Pub/Sub APIs fines how a system processes data. Pub/Sub (Publisher/Subscriber) is an infamous messag- ing design pattern. When it debuted in 1987 it solved loose decoupling of servers. The earliest distributed systems used to send events between servers using Pub/Sub, but all within reliable networks. This pattern operates by publishing API 365 Workshop: messages on a topic (or channel) within a message bus, Building APIs and managing and subscribers can listen for events based on those top- them with serverless, containers, ics. Pub/Sub is over 30 years old now. The design makes API gateways and Azure no attempt to define semantics around ordering or Dr. Roberto Freato (Freelance) continuity (loss of connection). While it’s still a pat- tern used widely today, these are issues that must be The main focus of this workshop is to considered and addressed in our world of unreliable have a look into the API ecosystem connections. to approach the problem with good insights and to organize things to be Ably provides a pub/sub (Image 2) API following this reliable, available and to reduce the model but with guarantees around ordering, continuity, overall governance effort. idempotency, and more. www.apiconference.net 4 WHITEPAPER API Design event in an event-driven world triggers a chain of events which must be processed down- stream, extending through the entire data supply chain. All components in this sup- ply chain are reactive, rapidly responding to events and per- forming onwards processing. The time-bound, reactive nature of this data supply chain causes increased engi- neering and infrastructure complexity. And when com- pared to REST APIs following the request/response model, the complexity is inverted and put onto API producers rather than API consumers. However, that’s a conver- sation beyond the scope of this article. But it’s something we’ve spoken about before at API Image 3: Ably’s unified push notifications API And so we arrive at Realtime APIs It simply states a system should be responding/re- Streaming, Pub/Sub, and Push are all patterns that can acting to events as they happen. Streaming, Pub/Sub, be delivered through event-driven architecture. But the and Push are all messaging patterns that can delivered outcomes of streaming, push, pub/sub, and event-driven through an event-driven architecture. As such they can are all realtime functionality. They’re all means of get- all fall under the umbrella of event-driven APIs. ting data from producer to subscriber in the shortest Most of us know what an event-driven API is but possible time. So we can call them realtime APIs. an example never hurts. Unlike traditional request/re- But you can’t have a realtime API unless it’s delivered sponse APIs where data is requested, event-driven APIs as an event-driven, pub/ sub, streaming, or push API. push data from a producer to a consumer. They can be Hence how we’ve settled on the umbrella term of real- quite simple or very complex. time API to encompass them all. In the above example a retailer needs an event-driven architecture to respond to things as they happen. An Navigating complexity and fragmentation As you can see, it’s easy to use these terms interchangea- bly despite them all being different. The realtime ecosys- tem is still maturing and changes all the time. Eventually we believe developers and organizations will naturally come to a shared and standardized terminology for APIs Do we have a Packaging designed to deliver real time functionality. Problem? Until then, the Realtime API Family helps us to reduce complexity when we explain our cloud infrastructure Maxim Zaks (Self Employed) and APIs to the developer community, our users, and our potential customers. Designing an API is a complex ende- avour. It contains of many details and considerations. One of such conside- rations is: How do we package data we are sending around? In this talk, we Kieran Kilbride-Singh (Product Marketing Manager) will discuss different packaging techniques like CSV, is writer and marketer with enough technical know- XML, JSON, Protocol Buffers and FlatBuffers. We will how to be dangerous in a GitHub repo. He first flexed evaluate the benefits and pitfalls those techniques his fingers writing about topics like interoperability in embedded IoT devices. Nowadays he thinks and bring to the table. writes about realtime infrastructure, event-driven APIs, and next-generation connected experiences for Ably.io. www.apiconference.net 5 WHITEPAPER API Development How to avoid common API misconceptions and the resulting disillusionment What APIs Can and Cannot Do APIs sometimes are perceived as having almost magical properties: Since they are so essential to digital transformation, and all of our digital experiences to- day are powered by APIs, the conclusion is that once you have and use APIs, all the wonders of digital transformation are simply going to happen.
Recommended publications
  • Open Source Used in Influx1.8 Influx 1.9
    Open Source Used In Influx1.8 Influx 1.9 Cisco Systems, Inc. www.cisco.com Cisco has more than 200 offices worldwide. Addresses, phone numbers, and fax numbers are listed on the Cisco website at www.cisco.com/go/offices. Text Part Number: 78EE117C99-1178791953 Open Source Used In Influx1.8 Influx 1.9 1 This document contains licenses and notices for open source software used in this product. With respect to the free/open source software listed in this document, if you have any questions or wish to receive a copy of any source code to which you may be entitled under the applicable free/open source license(s) (such as the GNU Lesser/General Public License), please contact us at [email protected]. In your requests please include the following reference number 78EE117C99-1178791953 Contents 1.1 golang-protobuf-extensions v1.0.1 1.1.1 Available under license 1.2 prometheus-client v0.2.0 1.2.1 Available under license 1.3 gopkg.in-asn1-ber v1.0.0-20170511165959-379148ca0225 1.3.1 Available under license 1.4 influxdata-raft-boltdb v0.0.0-20210323121340-465fcd3eb4d8 1.4.1 Available under license 1.5 fwd v1.1.1 1.5.1 Available under license 1.6 jaeger-client-go v2.23.0+incompatible 1.6.1 Available under license 1.7 golang-genproto v0.0.0-20210122163508-8081c04a3579 1.7.1 Available under license 1.8 influxdata-roaring v0.4.13-0.20180809181101-fc520f41fab6 1.8.1 Available under license 1.9 influxdata-flux v0.113.0 1.9.1 Available under license 1.10 apache-arrow-go-arrow v0.0.0-20200923215132-ac86123a3f01 1.10.1 Available under
    [Show full text]
  • Easybuild Documentation Release 20210907.0
    EasyBuild Documentation Release 20210907.0 Ghent University Tue, 07 Sep 2021 08:55:41 Contents 1 What is EasyBuild? 3 2 Concepts and terminology 5 2.1 EasyBuild framework..........................................5 2.2 Easyblocks................................................6 2.3 Toolchains................................................7 2.3.1 system toolchain.......................................7 2.3.2 dummy toolchain (DEPRECATED) ..............................7 2.3.3 Common toolchains.......................................7 2.4 Easyconfig files..............................................7 2.5 Extensions................................................8 3 Typical workflow example: building and installing WRF9 3.1 Searching for available easyconfigs files.................................9 3.2 Getting an overview of planned installations.............................. 10 3.3 Installing a software stack........................................ 11 4 Getting started 13 4.1 Installing EasyBuild........................................... 13 4.1.1 Requirements.......................................... 14 4.1.2 Using pip to Install EasyBuild................................. 14 4.1.3 Installing EasyBuild with EasyBuild.............................. 17 4.1.4 Dependencies.......................................... 19 4.1.5 Sources............................................. 21 4.1.6 In case of installation issues. .................................. 22 4.2 Configuring EasyBuild.......................................... 22 4.2.1 Supported configuration
    [Show full text]
  • Advanced Model Deployments with Tensorflow Serving Presentation.Pdf
    Most models don’t get deployed. Hi, I’m Hannes. An inefficient model deployment import json from flask import Flask from keras.models import load_model from utils import preprocess model = load_model('model.h5') app = Flask(__name__) @app.route('/classify', methods=['POST']) def classify(): review = request.form["review"] preprocessed_review = preprocess(review) prediction = model.predict_classes([preprocessed_review])[0] return json.dumps({"score": int(prediction)}) Simple Deployments @app.route('/classify', methods=['POST']) Why Flask is insufficient def classify(): review = request.form["review"] ● No consistent APIs ● No consistent payloads preprocessed_review = preprocess(review) ● No model versioning prediction = model.predict_classes( ● No mini-batching support [preprocessed_review])[0] ● Inefficient for large models return json.dumps({"score": int(prediction)}) Image: Martijn Baudoin, Unsplash TensorFlow Serving TensorFlow Serving Production ready Model Serving ● Part of the TensorFlow Extended Ecosystem ● Used internally at Google ● Highly scalable model serving solution ● Works well for large models up to 2GB TensorFlow 2.0 ready! * * With small exceptions Deploy your models in 90s ... Export your Model import tensorflow as tf TensorFlow 2.0 Export tf.saved_model.save( ● Consistent model export model, ● Using Protobuf format export_dir="/tmp/saved_model", ● Export of graphs and signatures=None estimators possible ) $ tree saved_models/ Export your Model saved_models/ └── 1555875926 ● Exported model as Protobuf ├── assets (Saved_model.pb)
    [Show full text]
  • Bringing Probabilistic Programming to Scientific Simulators at Scale
    Etalumis: Bringing Probabilistic Programming to Scientific Simulators at Scale Atılım Güneş Baydin, Lei Shao, Wahid Bhimji, Lukas Heinrich, Lawrence Meadows, Jialin Liu, Andreas Munk, Saeid Naderiparizi, Bradley Gram-Hansen, Gilles Louppe, Mingfei Ma, Xiaohui Zhao, Philip Torr, Victor Lee, Kyle Cranmer, Prabhat, and Frank Wood SC19, Denver, CO, United States 19 November 2019 Simulation and HPC Computational models and simulation are key to scientific advance at all scales Particle physics Nuclear physics Material design Drug discovery Weather Climate science Cosmology 2 Introducing a new way to use existing simulators Probabilistic programming Simulation Supercomputing (machine learning) 3 Simulators Parameters Outputs (data) Simulator 4 Simulators Parameters Outputs (data) Simulator Prediction: ● Simulate forward evolution of the system ● Generate samples of output 5 Simulators Parameters Outputs (data) Simulator Prediction: ● Simulate forward evolution of the system ● Generate samples of output 6 Simulators Parameters Outputs (data) Simulator Prediction: ● Simulate forward evolution of the system ● Generate samples of output WE NEED THE INVERSE! 7 Simulators Parameters Outputs (data) Simulator Prediction: ● Simulate forward evolution of the system ● Generate samples of output Inference: ● Find parameters that can produce (explain) observed data ● Inverse problem ● Often a manual process 8 Simulators Parameters Outputs (data) Inferred Simulator Observed data parameters Gene network Gene expression 9 Simulators Parameters Outputs (data)
    [Show full text]
  • Advances in Electron Microscopy with Deep Learning
    Advances in Electron Microscopy with Deep Learning by Jeffrey Mark Ede Thesis To be submitted to the University of Warwick for the degree of Doctor of Philosophy in Physics Department of Physics January 2021 Contents Contents i List of Abbreviations iii List of Figures viii List of Tables xvii Acknowledgments xix Declarations xx Research Training xxv Abstract xxvi Preface xxvii I Initial Motivation......................................... xxvii II Thesis Structure.......................................... xxvii III Connections............................................ xxix Chapter 1 Review: Deep Learning in Electron Microscopy1 1.1 Scientific Paper..........................................1 1.2 Reflection............................................. 100 Chapter 2 Warwick Electron Microscopy Datasets 101 2.1 Scientific Paper.......................................... 101 2.2 Amendments and Corrections................................... 133 2.3 Reflection............................................. 133 Chapter 3 Adaptive Learning Rate Clipping Stabilizes Learning 136 3.1 Scientific Paper.......................................... 136 3.2 Amendments and Corrections................................... 147 3.3 Reflection............................................. 147 Chapter 4 Partial Scanning Transmission Electron Microscopy with Deep Learning 149 4.1 Scientific Paper.......................................... 149 4.2 Amendments and Corrections................................... 176 4.3 Reflection............................................. 176
    [Show full text]
  • An Evaluation of Tensorflow As a Programming Framework for HPC Applications
    DEGREE PROJECT IN COMPUTER SCIENCE AND ENGINEERING, SECOND CYCLE, 30 CREDITS STOCKHOLM, SWEDEN 2018 An Evaluation of TensorFlow as a Programming Framework for HPC Applications WEI DER CHIEN KTH ROYAL INSTITUTE OF TECHNOLOGY SCHOOL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCE An Evaluation of TensorFlow as a Programming Framework for HPC Applications WEI DER CHIEN Master in Computer Science Date: August 28, 2018 Supervisor: Stefano Markidis Examiner: Erwin Laure Swedish title: En undersökning av TensorFlow som ett utvecklingsramverk för högpresterande datorsystem School of Electrical Engineering and Computer Science iii Abstract In recent years, deep-learning, a branch of machine learning gained increasing popularity due to their extensive applications and perfor- mance. At the core of these application is dense matrix-matrix multipli- cation. Graphics Processing Units (GPUs) are commonly used in the training process due to their massively parallel computation capabili- ties. In addition, specialized low-precision accelerators have emerged to specifically address Tensor operations. Software frameworks, such as TensorFlow have also emerged to increase the expressiveness of neural network model development. In TensorFlow computation problems are expressed as Computation Graphs where nodes of a graph denote operation and edges denote data movement between operations. With increasing number of heterogeneous accelerators which might co-exist on the same cluster system, it became increasingly difficult for users to program efficient and scalable applications. TensorFlow provides a high level of abstraction and it is possible to place operations of a computation graph on a device easily through a high level API. In this work, the usability of TensorFlow as a programming framework for HPC application is reviewed.
    [Show full text]
  • Master Thesis
    Master thesis To obtain a Master of Science Degree in Informatics and Communication Systems from the Merseburg University of Applied Sciences Subject: Tunisian truck license plate recognition using an Android Application based on Machine Learning as a detection tool Author: Supervisor: Achraf Boussaada Prof.Dr.-Ing. Rüdiger Klein Matr.-Nr.: 23542 Prof.Dr. Uwe Schröter Table of contents Chapter 1: Introduction ................................................................................................................................. 1 1.1 General Introduction: ................................................................................................................................... 1 1.2 Problem formulation: ................................................................................................................................... 1 1.3 Objective of Study: ........................................................................................................................................ 4 Chapter 2: Analysis ........................................................................................................................................ 4 2.1 Methodological approaches: ........................................................................................................................ 4 2.1.1 Actual approach: ................................................................................................................................... 4 2.1.2 Image Processing with OCR: ................................................................................................................
    [Show full text]
  • Curriculum Vitae
    Diogo Castro Curriculum Vitae Summary I’m a Software Engineer based in Belfast, United Kingdom. My professional journey began as a C# developer, but Functional Programming (FP) soon piqued my interest and led me to learn Scala, Haskell, PureScript and even a bit of Idris. I love building robust, reliable and maintainable applications. I also like teaching and I’m a big believer in "paying it forward". I’ve learned so much from many inspiring people, so I make it a point to share what I’ve learned with others. To that end, I’m currently in charge of training new team members in Scala and FP, do occasional presentations at work and aim to do more public speaking. I blog about FP and Haskell at https://diogocastro.com/blog. Experience Nov 2017-Present Principal Software Engineer, SpotX, Belfast, UK. Senior Software Engineer, SpotX, Belfast, UK. Developed RESTful web services in Scala, using the cats/cats-effect framework and Akka HTTP. Used Apache Kafka for publishing of events, and Prometheus/Grafana for monitor- ing. Worked on a service that aimed to augment Apache Druid, a timeseries database, with features such as access control, a safer and simpler query DSL, and automatic conversion of monetary metrics to multiple currencies. Authored a Scala library for calculating the delta of any two values of a given type using Shapeless, a library for generic programming. Taught a weekly internal Scala/FP course, with the goal of preparing our engineers to be productive in Scala whilst building an intuition of how to program with functions and equational reasoning.
    [Show full text]
  • KNIME Deep Learning Integration Installation Guide
    KNIME Deep Learning Integration Installation Guide KNIME AG, Zurich, Switzerland Version 3.7 (last updated on 2019-02-06) Table of Contents Introduction. 1 KNIME Deep Learning Integrations . 1 KNIME Keras Integration Installation. 2 Python Installation. 2 Installing the KNIME Keras Integration. 3 Extensions . 3 GPU Support. 4 KNIME TensorFlow Integration Installation . 4 Installation . 4 Advanced . 4 GPU Support. 4 KNIME Deeplearning4j Installation. 5 Installation . 5 GPU Support. 5 Known Issues . 5 KNIME Deep Learning Integration Installation Guide Introduction This document describes how to install the KNIME Deep Learning Integrations. These integrations bring deep learning capabilities to KNIME Analytics Platform, which allow you to read, create, edit, train, and execute deep neural networks within KNIME Analytics Platform. KNIME Deep Learning Integrations Three different deep learning libraries have been integrated: KNIME Keras Integration The KNIME Keras Integration utilizes the Keras deep learning framework to enable users to read, write, train, and execute Keras deep learning networks within KNIME. Furthermore, you can also build custom deep learning networks directly in KNIME via the Keras layer nodes. KNIME Tensor Flow Integration The KNIME TensorFlow Integration provides access to the powerful machine learning library TensorFlow* within KNIME. It enables you to read, write, train, and execute TensorFlow networks directly in KNIME. You can also convert your Keras networks to TensorFlow networks with this extension for even greater flexibility. * TensorFlow, the TensorFlow logo and any related marks are trademarks of Google Inc. KNIME Deeplearning4j Integration The KNIME Deeplearning4j Integration integrates the Deeplearning4j library into KNIME, which provides deep learning capabilities in Java. Within KNIME this means you can read, write, train, execute, and build Deeplearning4j networks.
    [Show full text]
  • From XML to Flat Buffers: Markup in the Twenty-Teens Warning! the Contenders
    Elliotte Rusty Harold [email protected] August 2018 From XML to Flat Buffers: Markup in the Twenty-teens Warning! The Contenders ● XML ● JSON ● YAML ● EXI ● Protobufs ● Flat Protobufs XML JSON YAML EXI Protobuf Flat Buffers App Engine X X Standard Java App Engine X Flex What Uses What Kubernetes X X From technology, tools, and systems Eclipse X I use frequently. There are many others. Maven X Ant X Google X X X X X “APIs” Publishing X XML XML ● Very well defined standard ● By far the most general format: ○ Mixed content ○ Attributes and elements ● By far the best tool support. Nothing else is close: ○ XSLT ○ XPath ○ Many schema languages: ■ W3C XSD ■ RELAX NG More Reasons to Choose XML ● Most composable for mixing and matching markup; e.g. MathML+SVG in HTML ● Does not require a schema. ● Streaming support: very large documents ● Better for interchange amongst unrelated parties ● The deeper your needs the more likely you’ll end up here. Why Not XML? ● Relatively complex for simple tasks ● Limited to no support for non-string programming types: ○ Numbers, booleans, dates, money, etc. ○ Lists, maps, sets ○ You can encode all these but APIs don’t necessarily recognize or support them. ● Lots of sharp edges to surprise the non-expert: ○ 9/10 are namespace related ○ Attribute value normalization ○ White space ● Some security issues if you’re not careful (Billion laughs) JSON ● Simple for object serialization and program data. If your data is a few basic types (int, string, boolean, float) and data structures (list, map) this works well. ● More or less standard (7-8 of them in fact) ● Consumption libraries for essentially all significant languages Why Not JSON? ● It is surprising how fast needs grow past a few basic types and data structures.
    [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]
  • Comparison and Benchmarking of AI Models and Frameworks on Mobile Devices
    Comparison and Benchmarking of AI Models and Frameworks on Mobile Devices 1st Chunjie Luo 2nd Xiwen He 3nd Jianfeng Zhan Institute of Computing Technology Institute of Computing Technology Institute of Computing Technology Chinese Academy of Sciences Chinese Academy of Sciences Chinese Academy of Sciences BenchCouncil BenchCouncil BenchCouncil Beijing, China Beijing, China Beijing, China [email protected] [email protected] [email protected] 4nd Lei Wang 5nd Wanling Gao 6nd Jiahui Dai Institute of Computing Technology Institute of Computing Technology Beijing Academy of Frontier Chinese Academy of Sciences Chinese Academy of Sciences Sciences and Technology BenchCouncil BenchCouncil Beijing, China Beijing, China Beijing, China [email protected] wanglei [email protected] [email protected] Abstract—Due to increasing amounts of data and compute inference on the edge devices can 1) reduce the latency, 2) resources, deep learning achieves many successes in various protect the privacy, 3) consume the power [2]. domains. The application of deep learning on the mobile and em- To make the inference on the edge devices more efficient, bedded devices is taken more and more attentions, benchmarking and ranking the AI abilities of mobile and embedded devices the neural networks are designed more light-weight by using becomes an urgent problem to be solved. Considering the model simpler architecture, or by quantizing, pruning and compress- diversity and framework diversity, we propose a benchmark suite, ing the networks. Different networks present different trade- AIoTBench, which focuses on the evaluation of the inference offs between accuracy and computational complexity. These abilities of mobile and embedded devices.
    [Show full text]