This course will focus on the fundamental systems research that enabled and continues to power modern cloud computing. We will cover the production systems that run within data centers operated by large cloud companies such as Google, Microsoft and Amazon, as well as the ground-breaking academic research that paved the way for these systems. Technically, we'll focus on the abstractions and mechanisms required to build online services that are scalable, highly available, durable, and consistent. We'll cover the entire stack, ranging from single-machine systems to protocols for distributing and replicating data within and across data centers. By the end of the course, students should be able to critically read a systems paper; distill, apply and improve upon key ideas from it; and write a short paper describing their own systems project.

Prerequisites: Equivalent of CPSC 323, or permission of instructor.

Syllabus: The course will be almost completely centered on a reading list of papers from top systems conferences (e.g., SOSP, OSDI, NSDI, FAST). These will span the last two decades of systems research and include seminal papers from academia (e.g., the Log-structured Filesystem, FAWN, Chain Replication), as well as state-of-the-art projects from industry (Google's papers on GFS/Spanner/Chubby, Amazon's Dynamo, Microsoft's Azure Storage, etc.). In each class, we will discuss one paper (sometimes two); these will either be presented by the instructor or by students. Your grade will be based on:

• summaries of each paper with strengths and weaknesses (2-3 paragraphs): 20%

• paper presentations: 20%

• a final project with a write-up: 50%

• class participation: 10%

Project details and timeline: There's a 1-2 page survey paper due by class on 3/8 that should contain a compact description of your project proposal (1 or 2 paragraphs), a project plan / timeline (a bullet list of what you expect to implement by when), and a survey of related work (which will end up fitting verbatim into your final report). For example, if your proposal involves implementing a new kind of replication protocol, your survey will describe existing replication protocols. For PhD students, the project should make a research contribution (e.g. a new replication protocol). For undergraduates and masters, they can either make a research contribution or take the implementation option and build a practically useful system (e.g. implement an existing replication protocol). You can optionally form groups of 2 people for the project.

Project topics: For the research route, read papers at recent conferences for research topics, or think about offshoots of the paper we read in class. For the implementation route, one approach is to pick any of the systems we have talked about in class (LFS, epidemic replication, Paxos, etc.) and build an implementation of it. The report should talk about how HW and applications have changed since the paper was published and what (if any) changes you had to make to the design as a result. User-space implementations are fine, but implementations are almost always preferred over simulation. If you are building a storage system, bytes must hit the storage medium; if you are building a distributed protocol, bytes must travel over the wire.

Paper reviews: Reviews are due via classesv2 before the class on which the paper is being discussed. PDF is preferred with your name and the class date on top.

Important dates:

• TBD

Administrative information:
Instructor: Mahesh Balakrishnan
Location and Timings: AKW 400, MW 1PM - 2:15 PM.
TA: Bo Hu.