CS 260r Projects and Close Readings in Software Systems
2019: SERVERLESS COMPUTING

MW 10:30–11:45, Pierce 213
Eddie Kohler

What is CS 260r?

• Research seminar in computer systems
• Relatively informal
• Different topic each time
• We read papers
• We may do a couple assignments
• We do research
• We learn together

Why read research papers?

• Learn currently hot topics in systems research and important past research
• Learn how to read research papers
  
• Learn how to do research
• Prepare for graduate school and other lifelong learning

Why do research?

• A major outcome for this course is your final research project
• A great project could be accepted for publication to a workshop or research conference—but what happens is up to you
• Group projects are OK
• Many projects will involve implementation or measurement
• A good research survey is also acceptable

Expectations

• Read papers before they are discussed in class
  • Each day will have a short survey question you should answer before class
• Present at least one paper
• Engage actively in class discussion
  • Attendance matters
• Complete assignments
• Project

Serverless computing

Cloud computing

Cloud computing

Utility computing

Utility computing

“…computing may someday be organized as a public utility just as the telephone system is a public utility…. Each subscriber needs to pay only for the capacity he [sic] actually uses, but he has access to all programming languages characteristic of a very large system. … Certain subscribers might offer service to other subscribers. … The computer utility could become the basis of a new and important industry.”

Utility computing

• It’s the 1960s, good computers are super expensive, almost no one has one
• But the expensive hardware stands idle most of the time
• Want to make the power of good computers available to everyone
• Technical issues around pricing, security, and management

Decline of utility computing

• Good computers get super cheap
• Even small organizations can run their own computing utilities
• Security and management issues are hard to overcome

Sidebar: ~visionary~

“My recollection is that I proposed the [space fountain] scheme [an enormous tower of hollow pipes and magnets, through which tiny iron rings are sent up to space for some reason] in 1982…. The work wasn’t published, because of difficulties with stability and protecting the tower from space junk. … Our scheme involved circulating the magnetic rings and getting almost all the energy back when the rings came down. The tower was to have been built with its base on Baker Island.”

Networks and big data

• Good computers get cheaper and cheaper, but large networks of good computers require big iron: switches, routers, cabling, disk farms, associated administration
• Enormous data processing requirements

Virtualization

• In the 1990s, and academic research innovations lead to a rebirth of virtual machine technology
• Goal: support operating system development on new hardware architectures
• SimOS (full-machine simulator) → DISCO → VMware
• Full machine image can be abstracted with high performance

Cloud computing

1. Largely-automated hardware management for networked computers
2. “shared pools of configurable computer system resources and higher-level services that can be rapidly provisioned with minimal management effort, often over the Internet” [Wikipedia]
3. Amazon Web Services

Cloud computing enabling technologies

• Virtual machines (different tenants can share hardware safely)
• Network address translation (breaks linkage between a physical computer and a network endpoint)
• Efficient block-based network storage (quicker, easier failover)

Cloud computing issues

• Faster provisioning, but still basically machine-based
• Automated hardware management is not enough

Serverless computing

1. Largely-automated operating system software management for networked services
2. “a cloud-computing execution model in which the cloud provider runs the server, and dynamically manages the allocation of machine resources” [Wikipedia]
3. Amazon Lambda

Serverless computing enabling technologies

• API virtualization (high-level triggers for calling functions/starting new computations)
• Container technologies (OS-level virtualization)
• Advanced provisioning (black box)

The dream

Serverless computing issues

• Latency
• Flexibility
• Opacity

Some project ideas

• Use serverless computing to improve an important application
• Measure current serverless computing offerings
• Analyze serverless computing platforms for security
• Develop operating system and software features specialized for serverless computing

Next time

• “The Computer Utility and the Community,” by Robert M. Fano
• “Serverless Computing: One Step Forward, Two Steps Back,” by Joseph M. Hellerstein et al.
• Thereafter: virtual machine technologies, NAT, eBPF, containers, …