Workload Automation for Cloud Migration and Machine Learning Platform

Paris, Sophia Antipolis, London, San Jose USAParis, Sophia Antipolis, London, San Jose USA, Montreal CA
2 Innovative Solutions for
1. Workload Automation for Cloud Migration
2. Data Science and Machine Learning Platform

ProActive Workflows & Scheduling (PWS)
For Gartner identified market of
“Workload automation and job scheduling” including Big Data & ETL
PRODUCT 1:

Main advantages of the solution
Technical differentiators
Automation, Orchestration, Scheduling
ActiveEon is an Open Source software editor providing Orchestration, Scheduling, IT and
Business Process Workload Automation for the multi-cloud era. Activeeon solution features
comprehensive resilient workflows at scale, scheduling & meta-scheduling, as well as an
integrated resource manager.
 Eases cloud migration
 Important savings on cloud expenses
 Time-to-value, speed of development
 Speed of execution
 Multi-system and multi-source data
aggregation
 Complex application scheduling,
distributed computing
 Cloud-ready: integrated resource
manager with elastic cloud scalability
 Scheduling and meta-scheduling of any
type of application
 Multi-language workflows at scale,
comprehensive and resilient
ProActive Workflows & Scheduling
Addresses Gartner identified
“Workload automation & job
scheduling” market, including
Big Data & ETL
Some customer references
www.activeeon.com

Workflows:All kinds & All Resources
</>
API
Scheduler & Meta-Scheduler
Resource Manager
Clusters
LSF
ProActive Workflows
Fault Tolerance
Cloud bursting
Resource agnostic

AE Architectural Vision
Studio: Easy to Use Workflow Designer,
with Palettes and Presets + APIs
PROACTIVE STUDIO
PROACTIVE
RESOURCES MANAGER
PROACTIVE SCHEDULER
PROACTIVE
CLOUD AUTOMATION
Automation Dashboard: PaaS, Job
Planner, Cloud Watch, Analytics
Scheduling & Meta-Scheduling: on
all Infra for Legacy, Cloud & Hybrid
Resource Manage: “Infrastructure as
Code”, Cost + Resource saving
Studio:
Automation Dashboard:
Scheduling & Meta-Scheduling:
Resource Manager:

Packaged Solution
Open Source
Activeeon Solution
Workflow
Studio
Intuitive, multi-
language, open
Error management, Job
Planner, Event Base
Schedule &
Control
Resource
Management
Hybrid, multi-cloud,
auto-scaling
Monitor
Real-time job and
resource monitoring

Comprehensive Scheduling for Cloud Era
All kinds of Scheduling and Workloads:
Pure Workload with high throughput
Time-Based
Event-Driven
Cloud-Centric: Fitted for the Cloud
On-Prem Physical & Virtual, Docker, All Clouds, Edges
Routing for Hybrid Clouds
Data Transfer
Capacity Control
Provisioning of Extra Capacity

Automation Dashboard - Catalog
Workflows stored in buckets in the Catalog
RBAC support for each bucket / Users can share workflows and templates
Keep track of the revisions with a versioning feature integrated

Job Planner
DefineCalendars AssociateWorkflowstoCalendars VisualizeExecutionPlanning
Manage recurring Jobs
Forecast and check future
Executions
Control recurring jobs from one
endpoint
Schedule Exceptions through
Exclusion Calendars &
Inclusion Calendars
Module 1:

Job Planner – Interactive GANTT Monitoring

Cloud Watch: Monitor/TriggerActions
Event-Based Scheduling
Module 2:

Cloud Automation: On-demand PaaS
On-Demand PaaS Services with full Life-Cycle Management
Module 3:

Scalable & Elastic resources
Incremental resource
deployment
100% resources
usage, no waste
Smart scale down
Provides cloud computing power according to your needs.
Minimize costs by deploying VMs only when needed (configurable
load factor). Never exceed your budget (min/max VMs threshold).
Smart and fully configurable elastic policy. Shutdown unused
VMs whenever it's possible. Prevent time-consuming re-
deployments by adjusting idle nodes’ release delay (avoid scale
up/down cycles).

Automatic GanttAnalysis to optimize your
Cloud Usage, and save on Cloud Expenses

PWS Customers – Cloud Adoption
AE technology is key for customers migrating to Clouds in all Verticals

Some Major Customers
Telco & IT Bio Tech
& Health
FinanceEngineering Aeronautics Energy
& Space
Some Partners:
Media
Distribution
Government
IoTCosmetics

Redistribute for free the products of Sentinel satellites, S1A, S1B, S2A and
S2B, S3A and S3B from COPERNICUS, the European system for the Earth
monitoring.
Multi-sensor (radar, optical, etc.), High frequency, long term project.
1 PB in 20 years and 7 PB in 2 years! 10 TB/day
PEPS: Plateforme d’Exploitation
des Produits Sentinel

Situation
Make Sentinel data available to the greatest number and
encourage the development of applications using them (agriculture, maritime field...)
Solution
Proactive Solution provided by ActiveEon to execute on Azure in hybrid mode
allows enhancing PEPS data and making them available to API providers :
• Multi-Cloud Ecosystem Platform
• Remove complexity for Data Scientists
• Provide Cloud performance
Benefits
• Faster execution, Optimisation of On-Prem ressources & Clouds,
• Easier to use by end-users
Space & Image Processing

PEPS: Sentinel Satellite ImageAnalysis
ProActive task
ProActive nodes
Key points to choose AE:
OnPrem & Clouds with a unique Task definition
Single Interface for Job Management
Out-of-the-box effective error management
Product integration with Docker
Business tasks can trigger VM provisioning on
Azure Cloud

L&G a leading multinational finance and insurance company with headquarters in London
Situation
European regulations: Solvency II, Basel III, etc.
Transform legacy system and embrace Cloud computing
Solution
Activeeon ProActive and migration to the Cloud have enabled
faster and more reliable execution:
• Cloud bursting
• Error management
• Prioritization
Benefits
From 18 hours to 2 hours for priority reports
Agile development with an objective of 4,000 cores
 $1.2m / year committed spent on Cloud
Finance
Time
64VMs,eachwith16vCPUs

Finance (Solvency Risk) on MicrosoftAzure
Time (hours)
105 160
1024Workers
Profile Results
1024Workers
Profile Results
I/O intensive tasks
Aggregation &
Reporting
Low Priority
CPU-Intensive tasks
Risk Watch Simulation
Medium Priority
CPU-Intensive tasks
High Priority
CPU-Intensive tasks
“More Value, Faster”
Full computation without intermediate result
High Priority
Results
Full Report
2H 5H
Batch optimization with ProActive:
• Enforce strong priorities
• Optimal compact execution
• Start tasks as early as possible
• Pipeline and co-allocate
• CPU-intensive with I/O intensive
tasks
18H
Achievements & Benefits:
High Priority Results in 2 H instead of 18 H !
Compacting Executions for savings on the Cloud
What Legal & General said about Batch optimization with ProActive:
“It enforces strong priorities, optimally compacts execution, starts tasks as early as
possible, pipelines and co-allocates CPU-intensive with I/O intensive tasks.”
32VMsorMachines
32VMsorMachines
Profile Results without ProActive Profile Results with ProActive

Solvency : Executing onAzure,
scaling to 1024 CPUs
 Workload that executes every day!
1 App: at least 1.2 M€/Year Azure Expenses
Customer is willing to scale up to 4 000 CPUs

Komatsu is a Japanese multinational corporation.
It manufactures construction, mining, industrial and military equipment.
Situation
ActiveEon Orchestrates on Cloud execution over hot and cold storage for Streaming and Batch Analytics
 1,200 tasks executed per hour
Solution
Activeeon has enabled control over scheduling and execution:
• Error Management – Notification, Automated Recovery
• Job Planner
• Distribution & Parallelization
Benefits
• Reliable execution to orchestrate multiple services and resources
• Provide consistent results and KPIs to end users and BI Tools
IoT
ActiveEon allowed to migrate
from AWS to Azure

Data from
Mining
Machine
Sensors
Health and
performance
of machines
Schedule data analytics
hourly or on events
Data
Real-Time
Control &
Optimizations
IoTAutomation in the Cloud
Data
Processing
on Premises &
in the Clouds
ActiveEon allowed to migrate
from AWS to Azure

UK Ministry of Interior is using ActiveEon for 2 critical applications:
• Visa Delivery Process, and
• Big Data & Analytics platform for Crime Reduction (HODAC).
Situation
25 different sources of Data.
uild a consolidated Data Lake and analytics platform to be used for many Home
Land security applications.
Solution
ActiveEon used as the central Orchestrator to Schedule and Meta-Schedule all the
Big Data, ETL, Analytics, Machine Learnigs software appliance of the platform
(Hadoop, SAS, TIBCO Spotfire, Python, Anaconda, GreenPlum, ElasticSearch, …).
Gov.: UK Ministry of Interior

Meta-Scheduler, Big Data ETL + new ELT
Execution
prod
Analytical
prod
Staging Dev
Virtualized Infrastructure using Docker
4 Thousands of physical cores
Main Benefits
Central Orchestration Tool
Workflow Expressiveness:
universal & comprehensive
Management of Security for
highly sensitive environments
Management of Resources for all appliances
(SAS, GREENPLUM, TIBCO, …)
« The only
solution capable
to Schedule any
Big Data
Analytics, mono-
threaded, multi-
threaded, multi-
core, parallel and
distributed »
Cap Gemini Lead
Engineer for
Home Office
EC2,
RDS DB
instance

CHALLENGES
Process 500 terabytes per year
Flexibility and enabler of interoperability
between heterogeneous services
Job affinity with data location
Transfer sensitive data to the cloud for
processing
RESULTS
Efficient metagenomics pipeline
Granular compute management
User friendly system for maximum utilization
Secure transfers
Simple workflow process definition
Workflow model and data management
Compute migration from on-prem to the cloud
MAIN DRIVER
REQUIREMENTS
Guidance and support to achieve high
performances
Fit in hybrid architecture multiplatform
Integration with R
FlexLM support (licenses manager)
Remote Visualization for interactive tasks
COMPANY PROFILE
Industry: BioTech
Product: Metagenomics
Clinical Research,
moving into Hospitals for Clinical Medicine

Hybrid CloudArchitecture Overview
Web Portal and
Integration with
Scientific tools
Total
DNA QC/Library preparation
Proton/Illumina
Sequencing
1.3PB
DataBase
1TB / Sequence
Analysis
Windows
Cluster 1
192 cores
Linux
Cluster 2
366 cores
Pre, Post Processing of Data Analysis
Flexibility, Speed of Analysis
Granular execution
Distribution for fast execution
Secure data transfer
Quantitative Metagenomics Platform
for gene profiling and statistical analysis
On-Prem Azure Cloud
CPU cores on demand
Cloud Bursting
REST / HTTPS
Secure data transfer

March 20th 2019, Round Table, Hybrid Cloud

INRATestimonial on Savings withAzure +ActiveEon
"When we invoice a sample to our partners, at the level of our IT we are between 5 and 10 €.
In the tests that we could do with ActiveEon and Azure, we are at 1 €, so a factor 5 to 10 won!",
Nicolas Pons, Research Engineer INRA, METAGENOPOLIS.

Cloud expenses optimization
Time to value, speed of development
Speed of execution
Multi-system, multi-data aggregationAutomation, Orchestration, Scheduling
ActiveEon is an Open Source software editor providing Orchestration, Scheduling, IT and
Business Process Workload Automation for the Multi-Cloud era. ActiveEon unique
solution features comprehensive resilient Workflows at scale, Scheduling & Meta-
Scheduling, integrated Resource Management.
Points for a successful
cloud strategy4
 Dynamic scaling based on workload
 Custom match between resources and jobs
 Focus on the job and not the pipeline
 Access larger resource pool for quick time to
result
 Get started in minutes
 Build complex workflows in minutes
 Any language and any resource
 For hybrid and multi-cloud
Cloud Migration withActiveEon

Machine Learning Open Studio (MLOS)
For Gartner identified market of
“Data Science and Machine Learning Platforms”, MLOps
PRODUCT 2:

Main advantages of the solution
Technical differentiators
Consistency, Repeatability, Scalability, Portability,
Loosely coupled architecture
Graphical workflow representation – Governance – Error management and alerting
– Any type of infrastructure – Integrated containers – Integrated pipeline models –
Python and Jupyter integration – Automation and parallelization of optimizations
with Auto ML pipelines – Incremental AI
 An open platform that allows to simplify,
accelerate and industrialize machine
learning
 Pipeline solution for machine learning
lifecycle automation
 Seamless execution at any scale in
production with any data source, on any
infrastructure
 Simple, portable, open and scalable
 Ready-to-use open source machine
learning and deep learning toolkits
 Connectors to cloud services
 Auto ML, incremental AI
 Execution over CPU, GPU, FPGA
 Versioning control and traceability
Machine Learning Open Studio
Addresses Gartner identified
“Data science & machine learning platforms”,
MLOps market
Some customer references
Automation of ML pipelines
on any infrastructure
www.activeeon.com
Data
preparation
Model
training
Model
optimization
Model
evaluation
Deployment Prod

All AI Libraries ready to be used
ML + DL
Full AI Pipeline
Auto ML & Incremental AI
Data Connectors
File Transfer
GPU & FPGA Support
Data Analytics
Job Analytics to accelerate AI developments
A Fully Open Platform

https://www.youtube.com/watch?v=mbrQxCf4lqM

Job Analytics – Select Jobs to Analyze

Job Analytics – Compare Var & Result

Monitoring Progress of your Jobs

Automate the whole pipeline
on any infrastructure
MLOS Value Proposition
Data
preparation
Model
training
Model
optimization
Model
evaluation
Deployment Prod
Simple
Portable
Scalable
Pre-built data connectors
Pre-built generic data prep. tasks
Pre-built containers
Pre-built pipeline templates
More …
Any infrastructure
Any resource: CPU, GPU, FPGA, …
Graphical workflow representation
Governance on workflow execution
Error management and alerting
Python SDK
Jupyter integration

End-to-end Industrialization of AI Pipeline
Automate and parallelize optimizations with Auto ML pipelines
Reuse manifest files for easy to use ML stackUse generic tasks Automate deployment pipelines
Data
preparation
Model
training
Model
optimization
Model
evaluation
Deployment Prod
Incremental AI and ML pipeline
Deploy everywhere, Share pipelines with everyone,
Share demanded resources, Control Versionning, Auto ML, Incremental AI

Benefits
• Auditable
• Reuse and standardize
• Automate
Use cases
• Build standard pipelines to transform Data or extract Features
• Automate Hyperparameters identification (Auto ML)
• Update production model regularly (Incremental Learning)
• Build generic tasks to improve code reusability
• Use Job Analytics to select best AI Models
Consistency / Repeatability

MLOS Value Proposition
Consistency
Repeatability
Create & automate
pipelines
Scaling
Portability
Openness
Loosely coupled
Run everywhere Standardize tasks

Direct execution from Jupyter withActiveEon Kernel
Execute seamlessly ML on local machine, on-premises on all HPC, on all Clouds and GPUs

Directly in Jupyter: Submit, See your Job executing, Get Results

For IT Department, &
High-Level Business & Application Owner:
• Simplify, Accelerate, Industrialize Machine Learning with
an Open platform.
• A pipeline solution that enable automation within the Machine
Learning dev lifecycle.
• Seamlessly execute at any scale in production with any data
source, on any infrastructure.
• Empowers data engineers and data scientists with a simple,
portable, open and scalable solution for machine learning
workflows.
• Save on Infrastructure expenses
MLOS Value Proposition -- ROI
TECHNICAL DIFFERENTIATORS
All Open Source ML & DL Toolkits
ready to be used
Connections to all AI Cloud Services
Open: Bring your own Toolkits
Auto ML
Incremental AI
Execute on CPU, GPU, FPGA
Version Control & Traceability

AI – Machine Learning Customer Cases

Machine Learning and
Statistics on MRI Images
by

Resource Management
Scientists
• Provide new statistical models on demand
• Develop ML models and algorithms at scale
• Manage distributed ML model training
• Deploy solution everywhere
Clinicians
• Request additional diagnostic on demand
• Identify new patterns for prediction and
advices
Hospital - Human Brain Project
Clinicians
on-prem
Scientists

Machine Learning Analytics for
Aerospace

Each campaign generates large amount of
data on more than 2000 sensors
Data generation
Scheduler
Resource Manager
Fault Tolerance
Cloud bursting
Resource agnostic
Micro-service
Etc.
Local Cloud
Dataexposition
ProActive allows to
parallelize data
analytics and scale the
infrastructure to
optimize compute
times.
Tests
2 ML Models: a Streaming Version during the Test and a Post-Mortem for fine grain detections
It also allows to Industrialize, Manage and continuously improve the Models.
Machine LearningAnalytics for aeronautics

FEATURES
● Predictive Aircraft Engine Monitoring
● ~50 Gb of compressed data (~1 Tb of
uncompressed raw data) collected from sensors
installed on the aircraft engines
● 44 trials (partitioned on 81 files) where 20 are
labelled (41 files)
● A total of 2163 sensors from 36 types
● 50 geographic families
BENEFITS
● Defect prevention with predictive maintenance
● Anomaly detection with machine learning
● Monitor sensor health during engine tests
● Faster results with parallel execution of
machine learning workflows
https://mse238blog.stanford.edu/2017/07/jega/iota-internet-of-things-for-aerospace/
Engine ON OFF
Anomaly Detection onAircraft Engine Sensors

FEATURES
● Data analysis: data cleaning and merging
● Signal processing
● Big data workflows for automation of test scenarios
● Automatic detection of the different categories of
anomalies (out of order, noisy, axis inversion)
● Data visualization in-browser
BENEFITS
● Data fetching from many sources
● Extraction of relevant features
● Faster results with parallel parsing and processing
● Automatic anomaly detection with machine learning
● Extensible & Evolving solution using Workflow tasks
ProActive workflow for Anomaly
Detection for Satellite Signals
Data visualization
No Anomaly
Noise
Out of order
Axis inversion
phase amplitude
Anomaly Detection on Satellite Sensors

FEATURES
● Machine Learning and Data Analytics for
Power Networks with Elasticsearch Logstash
Kibana (ELK)
● Anomaly Detection on Power Networks
● Peak Demand and Electricity Consumption
● Data Fusion, Clustering, Visualization
BENEFITS
● Code-free data analytics
● Efficiently real-time data stream processing and
analytics
● On-demand PaaS
Demo https://www.youtube.com/watch?v=dw65iADgCgc
Machine Learning and DataAnalytics for Power Networks

Deep Learning forAnomaly Detection in
Satellite Manufacturing
FEATURES
Detection of wires defect on a set of images
from production line using Deep Learning
Deep Learning on images of wires: occlusion,
variation, noise, grayscale, semantic analysis
Detection of defaults using a pre-defined wire
model and computing a distance measure
Workflows for model training and prediction for
parallel execution
BENEFITS
Automatic detection of defaults in hybrid
circuits manufacturing
Higher precision of Machine Learning results
Faster results with parallel execution of
machine learning workflows
Workflows can be used for other applications
Faulty wires come out in red

Big DataAnalysis forAutomatedAnomaly
Tracking in Satellite Communication
FEATURES
Data analysis: checking packets number of service
telemetries, order and type
Incident evolution forecasts
Big data workflows for automation of Test Scenarios
Automatic detection of remote controls that didn’t
receive expected telemetries
Data visualization in browser
BENEFITS
Automatic and early detection of defaults via trends
analysis of test results
Engineering process improvement: margin assessment,
robustness analysis, model elaboration based on actual
behaviors
Workflows allowing to accelerate treatments of fast-
growing test data amounts
Data fetching from many sources
ProActive workflow for service
telemetries verification
Visualisation of anomalies

ProActive workflow to predict banking clients default in the next 12 months.
FEATURES
● Predict banking clients default
● Detect fraud in financial payment services using
machine learning algorithms
BENEFITS
● Deal with unstructured financial data
● Parallel data loading and training
● Deal with a large volume of data
● Privacy and security
Predict banking client defaults in the next 12 months.
PredictiveAnalytics in Finance

FEATURES
● Introduce Artificial Intelligence technologies in the
aerial/satellite image chain
● Give them autonomy, opening up many
opportunities for remote sensing with a capacity
for reaction
● Creation of large database of labeled aerial
images dedicated to learning
● Evaluate performances on concrete settings (use
cases, data, hardware)
BENEFITS
● A collaborative platform to design, evaluate and
share ML algorithms
● Direct access to diverse hardware (different GPU,
FPGA, Clouds, On-premise)
● On-demand scalability for diverse experiments
ML for Image Analysis Onboard Satellites

Digital transformation for manufacturing
BENEFITS
Reduce the distance between the virtual and the
manufacturing process
Take advantage of digitalization in the machine tool
field for intelligent manufacturing and more efficient
production
FEATURES
Cloud-based big data analytics during
machining
Optimization of machining parameters using
workflows
Process simulation and optimization tools
Physical measurements and monitoring
Virtual / real part model correction
Use of AI
TARGETED SECTORS
Manufacturing, automotive, aerospace
Cloud processing services in manufacturing
END USERS

FEATURES
● Find ships directly on satellite images as
quickly as possible by semantic segmentation
using Deep Learning
● Semantic segmentation on binary images of
ships: objects from the same class with large
variability in terms of scale, pose, viewpoint
and background
BENEFITS
● Automatic detection of ships to support the
maritime industry to increase knowledge,
anticipate threats, trigger alerts, and improve
efficiency at sea
ProActive workflow for ship detection on satellite images
Input image
Output image
Non-public dataset: https://www.kaggle.com/c/airbus-ship-detection
Ship Detection on Satellite Images

Workflows for HPC multi-physics engineering
simulations in automotive and aerospace
BENEFITS
Thermal resistance for engine partsFEATURES
Parallel evaluation of optimal mesh size for
the best tradeoff between execution time
and result accuracy
Complex workflow management: monitoring,
scheduling and orchestration
Infrastructure management: on-premises and
cloud HPC
Data collection and processing
END USERS
Pollution levels in a district
Workflow for exploration of tradeoff
between execution time and result accuracy
DOMAIN: COMPUTATIONAL FLUID DYNAMICS (CFD) AND POST-PROCESSING TOOLS
Acceleration and Automation of
Design Analysis and Optimizations

Paris, Sophia Antipolis, London, San Jose USA @activeeon
contact@activeeon.com
+33 988 777 660
Automate Accelerate & Scale
10K Nodes, 20K Tasks, 1M Jobs

Workload Automation for Cloud Migration and Machine Learning Platform

Recomendados

Recomendados

Más contenido relacionado

La actualidad más candente

La actualidad más candente (20)

Similar a Workload Automation for Cloud Migration and Machine Learning Platform

Similar a Workload Automation for Cloud Migration and Machine Learning Platform (20)

Más de Activeeon

Más de Activeeon (6)

Último

Último (20)

Workload Automation for Cloud Migration and Machine Learning Platform