Advanced Topics in Computer Systems (Spring 2020)

Introduction and Course Overview

This meeting will be overview of the class and introduction of class members. Time permitting, we will talk about system performance.

Student 3-minute “My Systems Research” presentations.

Each class member will offer a 3-slide presentation of their current research focus.

Add presentation or link here

Sign up for debate roles

Great Debate 1: OS Abstractions

A fundamental question in systems is how system software creates effective abstractions enabling protected access to shared hardware resources. We will have two debates centered on HotOS papers - extensible kernels and virtual machines.

Great Debate 2: Threads vs Events

A critical element of all operating systems is how to handle concurrency. Fundamentally there have been two camps, threads vs events. Needham tried to put the questiojn to rest in 1978 with the Duality paper. But graphical user interfaces and then internet services and wireless embedded networks brought it back to the fore. The SOSP 2001 program committee is debated whether to reject the SEDA paper or to make it a keynote. The compromise was to accept it and have Roger Needham deliver a keynote adjacent to it. But two years later, some of the SEDA authors had evolved their views. You’ll want to read or at skim the optional references, but you don’t need to report on them.

Debate 3: Multicore OS

With the end of clock scaling we have experienced a “fourth wave” of multiprocessor design, MultiCore, where wee can expect the number of cores per processor to grow rapidly as chip density continues to grow. As the multi-micrioprocessor revolution of the 80’s drove BSD, this presents basic questions of scaling Linux vs rethinking the design, but now where cloud-based services are the driving workload, behooving us to revisit the contract between system and user level, represented in the 90’s by Scheduler Activations.

Debate 4: HLL role is OS Design

Throughout the histiory of computing, operating systems have represented the most demanding sort of complex, concurrency intensive, high-reliability software. This has given rise to numerous new language designs, and new languages have offered the potential to finally leap beyond C as the implementation language. Recent years have seen a renaissance in programming language design, including new systems languages, e.g, Go and Rust.

Scan on KV - MapReduce

As cluster computing took off post-search engine, Scan algorithms were rediscovered over a new storage model while handling unreliability.

Distributed Consensus in Practice

Paxos has dominated the vernacular of distributed consensus. Not only does it have its own checked history of whether it warranted publication, but the efforts to develop a more understandable and implementable replacement (Raft) were rejected three times before being published. These readings overlap with CS262A; we will take the opportunity to explore overcoming rejection in the pubication process.

Here are the reviews from the 3 rejected submissions of Raft, before being accepted and awarded Best Paper. The RAFT link provides the conference talk. Watch it before class.

Emergence of Planetary Scale Computing

As clusters, content-distribution networks (e.g., Akamai), and PVM took hold, but the Cloud gaining massive scale, global geographic scale, and general usage was economically inconceivable, visionary explorations of planetary scale computing emerged. The WebOS paper was rejected four times, but invited by a PC chair. Planetlab faced distainful reviews, but recieved multiple test-of-time awards a decade later. The Grid remained largely outside the systems research community.

Cloud Cluster Frameworks

Term projects in 2009, Mesos and Spark, brought an essentially level of generality over the opensource, Hadoop, and commercial state of the art. The “cluster level” systems design concepts were able to reach much further with the maturation of cluster technology and its “local level” systems building blocks.

NSDI

NSDI

CS262B Alums - Entreprenurialism and Graduate Systems

Societal-Scale Graph Processing

SysML - Cloud-scale Machine Learning Systems

Serverless

Planetary Scale Computing Systems Redux

EECS Faculty Retreat

Class time available to work on projects.

Spring Break

Spring Break

Debate: Post-Cloud Operating System Design

Network Function Virtualization

IFC and IPC

Time and Defensive Distributed System Design

Epidemic Routing and Gossip Protocols

Key Results - Student Presentations

Optional Project Discussions

How to be a good reviewer

You do not need to do write-ups, but please read these. We will have a discussion about the PC review process, what to expect from reviewers, and what expectations you should live up to when you take on that role. The optional is quite useful too.

No Class

Draft Reports due 4/29.

Failure and Disaster Planning

Draft reports distributed for review

RR Mock PC Meeting

5/7/20 RR