PA200 - Cloud Computing
Lecture 6: Cloud providers
by Petr Blaho and Ilya Etingof, Red Hat
In this lecture
• Cloud service providers
• IaaS/PaaS/SaaS and variations
• In the eyes of the user
• Pros&Cons
• Google Cloud walk-through
Cloud software vs cloud service
• Cloud software provider
• OpenStack, RedHat OpenShift, oVirt
• Many in-house implementations
• Cloud service provider
• Amazon EC2
• Microsoft Azure
• Google Cloud Platform
• RedHat OpenShift Online
• ...
What does CSP do?
A combination of:
• IaaS
• PaaS/Stackless/FaaS
• SaaS
Business use of CSPs
The hyperscalers:
What does IaaS CSP do?
• Abstracts away the hardware
• Operating system as a unit of scale
(before IaaS, hardware computer has been a unit of scale)
IaaS CSP business model
• Owns/rents physical DC infrastructure
• Owns/buys Internet connectivity (links, IX etc)
• Provides IaaS to end customers
• Serves other CSPs: PaaS and SaaS
• Base for multi-cloud CSPs
Typical IaaS offering
• Compute nodes (VMs)
• Virtual networks
• Bare metal nodes
• Managed storage (block, file systems)
• Instance-based scaling and redundancy
• Pay per allocated resources (instances, RAM, storage, traffic)
Example IaaS
• Amazon Elastic Compute Cloud (EC2)
• Google Cloud
• Microsoft Azure Cloud Computing Service
IaaS CSP differentiation
• Technically interchangeable
• Similar costs
• Customer is not heavily locked-in
Multicloud
Multiple cloud services under the single control plane
• Reduces dependence on a single CSP
• Balances load/location/costs
• Gathers resources
• Same deployment model (unlike hybrid cloud)
Examples of multicloud software
• IBM Cloud Orchestrator
• RedHat CloudForms
• Flexera RightScale
What does PaaS CSP do?
• Abstracts away the OS
• Containerized application as a unit of scale
PaaS CSP business model
• Owns or rents the IaaS
• Maintains the platform
• Maintains services, data collections etc.
Typical PaaS offerings
• Application lifecycle management
• Application runtime environment
• Managed services through proprietary APIs
• Container-grain scaling and redundancy
Example PaaS CSPs
• AWS Elastic Beanstalk
• Azure App Service
• IBM Bluemix (IBM Cloud)
• Google Kubernetes Engine, AppEngine
• Heroku
Example PaaS: Cloud Foundry
• Containerized applications
• Deployed in "buildpacks"
• Java, Python, ruby, .NET, Go, PHP run times
• External access through service broker
• Microservices marketplace
Higher-level PaaS services
• Serverless
• Functions (FaaS)
What does Serverless do?
• Abstracts away application runtime (no container anymore!)
• Assembled applications as a unit of scale
Example serverless: Google App Engine
• Python, Java, Go, PHP, Node.js run times
• Applications deployed through SDK
• Application is called on HTTP request, ends in 60 secs max
• Google Cloud Datastore access through GQL
What does FaaS do?
• Abstracts away language runtime
• Function as a unit of scale
FaaS
• Event driven applications
• Charged for run time, not instance sizes
• Scaling is event-driven, not instance-based
Example FaaS
• Amazon Lambda
• Google Cloud Functions
• IBM Cloud Functions
• Azure Functions
• Oracle Fn Project
Example FaaS: Google Cloud Functions
• Fully managed language runtime (Node.js, Python etc)
• Events via HTTP or Pub/Sub
• Pay per use
PaaS: managed services
• Big Data
• Cloud AI
• IoT
Example PaaS AI services
• Natural Language Text Analysis
• Translation API
• Vision API
• Video Intelligence
• Maps API
PaaS CSP differentiation
• Technically unique and incompatible
• Customer can be locked-in
What does SaaS CSP do?
• Managed application
• Subscription-based, no licensing fees
• No deployment phase
• User access from a thin client
• Online integration with company's IS
• Portals vs Mashups
SaaS benefits & challenges
Pro:
• Can be easier, cheaper or free to use
• Unlimited storage/processing capabilities
• Rolling upgrades
Contra:
• Data ownership
• Access reliability and latency
• Abrupt product EOL
SaaS applications
• Office and collaboration tools
• Payroll and HR systems
• Development and engineering software (CAD)
• Customer relationship management (CRM)
• Enterprise resource planning (ERP)
• Content management (CM)
• Geographic Information Systems (GIS)
Example SaaS
• Office365: tools for business
• Slack & Salesforce: collaboration tools
• Workday: HR tools
• ServiceNow: enterprise automation
• Shopify: e-shops
• Veeva Systems: pharmaceutical
Let's deploy an app
In Google Cloud:
• VM
• Container
• AppEngine
• Function
IaaS: Google Compute
1. Create and install VM instance
• Choose flavor, distro
1. Configure the instance
• Install language runtime and dependencies
• Install/configure DB etc.
3. Expose the VM & app to external traffic
• Setup virtual network
• Configure virtual router, public IP, DNS record
• Update local firewall
PaaS: Google Kubernetes Engine
1. Create a Docker container image
• Create Dockerfile
• Pack app & dependencies into container
• Upload container into the registry
2. Setup GKE
• Setup a cluster
• Create a pod of containers
3. Expose pod & app to external traffic
PaaS: Google AppEngine
1. Create the app.yaml file describing the app
• Specify language runtime
• Requested resources (CPU cores, RAM, storage sizes)
• Autoscaling boundaries
2. Deploy the source code
PaaS: Google Functions
1. Select function runtime properties
• Max RAM
• Trigger: HTTP, messaging
• Access URL
2. Upload source code
Recap: CSP's
• Hyperscalers offer IaaS + PaaS + SaaS
• PaaS & SaaS base on IaaS from hyperscalers
Recap: many forms of PaaS
• Containers
• Stackless
• FaaS
Recap: cloud can be sticky
• Vendors want to lock you in
• PaaS is especially sticky
Recap: choice of cloud
• Depends on technical requirements
• Desired degree of flexibility
• Overall cost of the scenario
Q&A
?