Serverless Computing

• Computing without servers?
• Running applications without the need to manage servers?
• Running functions instead of containers/VMs?
• Infinite scaling?
• The truth: no clear, agreed definition, i.e., no one really knows

One Perspective: How Cloud and Virtualization Evolved

Decreasing concern (and control) over stack implementation, Increasing focus on business logic

A Related Topic: Microservice

• A software architecture that develops an application as a suite of small services, each of which can be deployed and scaled independently
• When one (micro)service is in large demand, can scale it up
• Different (micro)services can be written and managed by different teams
• Changing one (micro)service will not affect the others

What is the essence of “Serverless Computing”?

• Management-free
• Autoscaling
• Only pay for what you use

What is Today’s Serverless Computing Like?

• Largely offered as Function as a Service (FaaS)
  • Cloud users write functions and ship them
  • Cloud provider runs and manages them
• Still runs on servers
• Have attractive features but also many limitations

Basic Architecture

…

BigQuery, …

…

AWS Lambda

• An event-driven, serverless computing FasS platform introduced in 2014
• Functions can be written in Node.js, Python, Java, Go, Ruby, C#, PowerShell
• Each function allowed to take 128MB – 3GB memory and up to 15min
• Max 1000 concurrent functions
• Connected with many other AWS services

Lambda Function Triggering and Billing Model

• Run user handlers in response to events: RPC handlers, triggers, cron jobs
• Pay per function invocation: no charge when nothing is run

Internal Execution Model

• Developers upload function code to a handler store (and associate it with a URL)
• Events trigger functions through RPC (to the URL)
• Load balancers handle RPC requests by starting handlers on workers
• Handlers and sandboxed in containers

…

Azure, GCP, …

…

Limitations of Today’s Serverless Offerings

• Difficult and slow to manage states
• No easy or fast way to communicate across functions
• Functions can only use limited resources
• No control over function placement or locality
• Billing model does not fit all needs