CS262A: Advanced Topics in Computer Systems

Fall 2026, UC Berkeley

Timing: Tuesday and Thursday, 12:30pm to 2:00pm

Location: The Gateway Building B1022

Instructor: Rishabh Iyer

GSI: Shubham Mishra

Course description: This is a graduate survey of Computer Systems covering a breadth of topics including, but not limited to, early/classical systems, file systems, transactions, distributed systems, security and AI systems. The objective is to get into the habit of reading systems papers, developing the ability to extract the key design insights, and being able to discuss the strengths and weaknesses of the arguments presented in the papers. Students are expected to lead and contribute to discussions in class, while working towards a group research project of publishable quality.

Grading:

• 40%: class participation (includes paper reviews).
• 20%: paper presentations.
• 40%: course project.

Paper Reviews: Please submit your reviews here. Reviews are due 24 hours before class.

Schedule

Aug 27 (Thu) — Introduction

• The UNIX Time-Sharing System (Slides)

Sep 1 (Tue) — Databases

• A History and Evaluation of System R (Slides)

Optional:

• The Design and Implementation of INGRES
• ARIES: A Transaction Recovery Method Supporting Fine-Granularity Locking and Partial Rollbacks Using Write-ahead Logging

Sep 3 (Thu) — File Systems

• A Fast File System for UNIX
• The Design and Implementation of a Log-Structured File System

Slides

Optional:

• Debates on Log structured file system: Seltzer's Paper, Ousterhout's critique, Seltzer's Response.

• Analysis and evaluation of Journaling File Systems

Sep 8 (Tue) — Transactions

• Granularity of Locks and Degrees of Consistency in a Shared Database
• Principles of Transaction-Oriented Database Recovery

Sep 10 (Thu) — OS Classics

• Microkernel Operating System Architecture and Mach
• Exokernel: An Operating System Architecture for Application-Level Resource Management

Sep 15 (Tue) — OS (II)

• The Multikernel: A New OS Architecture for Scalable Multicore Systems

• The eBPF Runtime in the Linux Kernel

Optional:

• SPIN - An Extensible Microkernel for Application-specific Operating System Services

Sep 17 (Thu) — Consensus

• Paxos Made Moderately Complex
• In Search of an Understandable Consensus Algorithm (Raft)

Optional:

• Paxos Made Simple

Sep 22 (Tue) — Distributed Storage

• The Google File System
• Bigtable: A Distributed Storage System for Structured Data

Sep 24 (Thu) — DHTs, KV Stores

• Chord: A Scalable Peer-to-peer Lookup Service for Internet Applications
• Dynamo: Amazon's Highly Available Key-value Store

Sep 29 (Tue) — Distributed Coordination

• Coordination Avoidance in Database Systems
• CRDTs: Consistency without Concurrency Control

Oct 1 (Thu) — Virtualization

• Disco: Running Commodity Operating Systems on Scalable Multiprocessors
• Xen and the Art of Virtualization

Oct 6 (Tue) — Cluster Management

• Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center

• Borg, Omega, and Kubernetes

Oct 8 (Thu) — Scheduling

• Lottery Scheduling: Flexible Proportional-Share Resource Management

• Dominant Resource Fairness: Fair Allocation of Multiple Resource Types

Oct 13 (Tue) — Datacenter Applications

• Web Search for a Planet: The Google Cluster Architecture
• The Tail at Scale
• Attack of the Killer Microseconds
• IX: A Protected Dataplane Operating System for High Throughput and Low Latency

Optional:

• Arrakis: The Operating System is the Control Plane
• Shinjuku: Preemptive Scheduling for $\mu$second-scale Tail Latency

Oct 15 (Thu) — Distributed Data Processing

• MapReduce: Simplified Data Processing on Large Clusters
• Spark: Cluster Computing with Working Sets

Optional:

• Apache Flink™: Stream and Batch Processing in a Single Engine

Oct 20 (Tue) — Serverless

• Firecracker: Lightweight Virtualization for Serverless Applications
• Unifying Serverless and Microservice Workloads with SigmaOS

Oct 22 (Thu) — Disaggregated/CXL

• LegoOS: A Disseminated, Distributed OS for Hardware Resource Disaggregation
• Tigon: A Distributed Database for a CXL Pod

Oct 27 (Tue) — Security & Privacy

• EROS: a fast capability system
• Private Web Search with Tiptoe

Optional:

• Making Information Flow Explicit in HiStar

Oct 29 (Thu) — Verification

• seL4: Formal Verification of an OS Kernel
• Verus: A Practical Foundation for Systems Verification

Optional:

• Using Crash Hoare Logic for Certifying the FSCQ File System

Nov 3 (Tue) — Parallel Programming

• Ray: A Distributed Framework for Emerging AI Applications
• Halide: Decoupling Algorithms from Schedules for High-Performance Image Processing

Optional:

• Erlang: Making reliable distributed systems in the presence of software errors

Nov 5 (Thu) — ML Programming Frameworks

• TensorFlow: A System for Large-Scale Machine Learning
• PyTorch: An Imperative Style, High-Performance Deep Learning Library

Nov 10 (Tue) — Pre-training

• ZeRO: Memory Optimizations Toward Training Trillion Parameter Models

• PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel

Nov 12 (Thu) — Post-training

• HybridFlow: A Flexible and Efficient RLHF Framework

• OpenRLHF: An Easy-to-use, Scalable and High-performance RLHF Framework

Nov 17 (Tue) — Inference (I)

• Efficient Memory Management for Large Language Model Serving with PagedAttention (vLLM)
• NanoFlow: Towards Optimal Large Language Model Serving Throughput

Nov 19 (Thu) — Inference (II): Multi-modal, MoT

• StreamDiffusion: A Pipeline-level Solution for Real-Time Interactive Generation
• M*: A Modular, Extensible, Serving System for Multimodal Models

Nov 24 (Tue) — Thanksgiving

• No class

Nov 26 (Thu) — Thanksgiving

• No class

Dec 1 (Tue) — AI for Systems

• Guest lecture; Readings TBD

Dec 3 (Thu) — Systems Support for Agents

• Guest lecture; Readings TBD

Dec 8 (Tue) — OSDI Deadline

• No class

Dec 10 (Thu) — Poster Day

• No class