Share
Embed
Quelle compétition est possible entre les 3 principaux clouders publics ?
Quelle compétition est possible entre les 3 principaux clouders publics ?
Laurent Grangeau
November 13, 2019
Transcript
41 slides
-
QUELLE COMPÉTITION EST POSSIBLE ENTRE LES 3 PRINCIPAUX CLOUDERS PUBLICS
? OLIVIER DUPRE - @OlivierDupreTec TONY JARRIAULT - @jarriaultTony LAURENT GRANGEAU - @laurentgrangeau -
#LifeAtSogeti Olivier Dupré Preparing AWS & GCP certifications Tony Jarriault
AWS Pro certified Laurent Grangeau Microsoft MVP Docker and AWS Certified -
Objective: Compete with our skills and knowledge to see more
clearly in the giant ecosystem of public clouds. -
AWS Historical cloud leader with EC2, S3 & SQS Started
in 2006 Azure Historical enterprise partner with Windows & Office 365 Started in 2010 with SQL GCP Challenger since 2011 Launched AppEngine in 2008 Launched GKE in 2014 Market Penetration -
Fear the rules • GDPR, Patriot Act, Cloud Act GCP
✔ No DC in France nor Africa ✔ No POP in Finland → using MPLS AWS & Azure (DC) ✔ France: Marseille & Paris ✔ Africa: Johannesburg & Cape Town Location -
AWS • VPC or VNet are regional Azure • Hub
& Spoke or Full Mesh topologies GCP • VPC is global ◦ Firewall rules based on tags/labelling Network Not transitive -
AWS & Azure • VPC or VNet are regional Azure
• Hub & Spoke or Full Mesh topologies GCP • VPC is global ◦ Firewall rules based on tags/labelling Network -
SLA • AWS & Azure ◦ No SLA on VM
◦ SLA on network / connectivity (99,99%) • GCP ◦ SLA on VM & Network / connectivity (99,99%) Quality / Availability HA & Fault Tolerance • AWS & GCP ◦ Distribute VM across AZ ▪ enhance availability ▪ enable HA or Fault Tolerance • Azure ◦ Availability Set do it for you -
No leader The three provide: • Block Storage at OS
level • File & Object Storage at application/API levels Object storage from Short Term → Long Term → Tape • New protocol created by AWS: S3 ◦ API became a standard de facto • Policies to migrate automatically to cheaper storage Storage : No limit Free to add data, cheap to store, expensive to retrieve Storage -
AWS & GCP •Network availability: 99.99% oStorage availability: from 99.0%
to 99.99% (depending on storage classes) AWS Refund up to 100% GCP Refund up to 50% •Storage durability: 11 9’s Means replicated at least 3 times SLO & SLA - Everything fails, all the time (Werner Voegels) Azure •Network availability: Starts at 95% oStorage availability: 98.0% to 99.99% (depending on classes) (economy !?) Azure Refund up to 100% oStorage durability: from 11 to 16 9’s -
Started with ECS + Fargate (firecracker) ✔ VPC & Subnet
isolation ✔ Multitenancy Late release of EKS --2018 in US Enables fully shutting down a cluster easily Optimizes cost CI/CD… needs custom integration with Lambda Allows different scaling pools with pain CaaS - AWS -
Came late with AKS (Oct 2017) and ACI (Azure Container
Instances) Max 400 nodes / 8 pools (GPU, CPU, …) No private cluster, no access to master, no HTTPS for API access on master Perfect integration with Gitlab & Azure DevOps integration channel: •Service Account, •Namespace •Environment CaaS - Azure -
Multi-pools Max 5000 nodes Autoscaling on K8S metrics CloudRun for
serverless Free of charge if free of workload Smoothly integrates with Gitlab.com & CloudBuild KaaS – GCP (K8S first contributor) -
AWS: App Mesh • Fargate + ECS & EKS only
• Runs outside of the cluster Azure: Service Fabric Mesh • capability added to Service Fabric • incompatible with K8S • Native integration with dev. pipelines • Azure SMI: brand new spec for SM GCP: Istio • Leader • Wider adoption • Works “virtually” everywhere • Runs inside the cluster -
AWS: 2014 Lambda Supported language: Node.js / Python / Java
/ C# / Go / Ruby / Powershell Azure: 2016 Azure cloud functions Numerous supported language: dotNet / PHP / Java / Javascript / Python / Typescript / Bash GCP: 2016 Cloud functions Few supported language: Javascript / Python / Go CloudRun Runs any image Serverless ❖ Deploy applications by running functions on-demand, without requiring an execution environment to be running constantly. ❖ No server maintenance -
Azure Stack in 2017 Azure Arc since November 2019 •The
leader : Greater coverage of services •Kubernetes (multi-cloud) •Infrastructure •Pricing : In previous mode Pay-as-you-go package in use •Operations done by you or service provider Hybrid - Azure -
GCP Anthos •Release in progress •Relies on VMWare •Implies minimum
workload ◦ Expensive ◦ Complex deployment •Container everywhere ◦ GKE ◦ Istio ◦ CI/CD •Marketplace Hybrid - GCP -
AWS RDS extended on VMWARE (vsphere) AWS Outpost When: late
2019 ◦To do What ? For the ELB / ECS / EKS / EMR /(big data) / RDS •Appliance including PaaS •VMWare or equivalent based •Unknown pricing •Operations done by AWS •Compliant Vmware API Hybrid - AWS -
AWS •Global Reach •Innovation •Long life reliable services •Wide range
of services •Very large set of ML/IA services •IaaS oriented Clouder Strengths AZURE •Hybrid solution •Easiest migration •Great for windows-based organizations •Dev & PaaS oriented GCP •Mastery container model •AZ eco-conscious option •ML research / Best algo for ML/IA •StackDriver •Dev & Container oriented -
Thank you for attending!!! OLIVIER DUPRE - @OlivierDupreTec TONY JARRIAULT
- @jarriaultTony LAURENT GRANGEAU - @laurentgrangeau -
Everyone knows these 3 major public clouds. But... what do
they really have in common? What are their major differences? Is the choice for one of the 3 a question of heart, available skills or technical abilities? -
It already takes more than a day to go through
each of these platforms. So going around the 3 in an exhaustive way during a talk is utopian. We will therefore get to the point and focus on the major services, the most used and for which the comparison is the most interesting. -
Beyond our certifications, all 3 of us are part of
the Cloud & Technos department of Capgemini Sogeti ATS and work necessarily, every day, in these environments. Without ever drawing a tight line between our playgrounds with the objective of challenging ourselves on a daily basis on the implementation of good practices. -
• Is that truly necessary to identify users and grant
different access? • Trust over control AWS •Cognito •Directory service •TrustAdvisor •1 out of 5 major pilars GCP •Directory Management •Cloud identity . Directory Service pour tous Azure •Active Directory •Windows based •Easily breakable IAM -
GCP •Stackdriver ◦All in one solution ◦Enables hot logging Azure
•AzureMon All in one solution? AWS •CloudTrail •CloudWatch •Xray Monitoring -
AWS • SageMaker • Comprehend • Lex • Polly •
Rekognition • Machine Learning • Translate • Transcribe • DeepLens • Deep Learning AMIs • Apache MXNet on AWS • TensorFlow on AWS ML / IA #Hashtag @twitter AZURE • Machine Learning • Azure Bot Service • Cognitive Services GCP • Cloud Machine Learning Engine • Dialogflow Enterprise Edition • Cloud Natural Language • Cloud Speech API • Cloud Translation API • Cloud Video Intelligence • Cloud Job Discovery (Private Beta) -
AWS • IoT Core • FreeRTOS • Greengrass • IoT
1-Click • IoT Analytics • IoT Button • IoT Device Defender • IoT Device Management IOT: Everywhere / Billion of devices AZURE • IoT Hub • IoT Edge • Stream Analytics • Time Series Insights GCP • Iot core