Fall 2026 Graduate Seminar

CS 395T: Hardware-Software Co-design for Cloud and Datacenters

Schedule and Readings

Back to Course Overview

Topic	Day	Readings
Course Introduction	Tuesday (08/25)	Course overview & Syllabus
Datacenter Workloads	Thursday (08/27)	Profiling a Warehouse-Scale Computer Profiling Hyperscale Big Data Processing A Cloud-Scale Characterization of Remote Procedure Calls (optional)
	Tuesday (09/01)	Accelerometer: Understanding Acceleration Opportunities for Data Center Overheads at Hyperscale DCPerf: An Open-Source, Battle-Tested Performance Benchmark Suite for Datacenter Workloads
Microservices + Software	Thursday (09/03)	An Open-Source Benchmark Suite for Microservices and Their Hardware-Software Implications for Cloud & Edge Systems µTune: Auto-Tuned Threading for OLDI Microservices Towards Modern Development of Cloud Applications (optional)
	Tuesday (09/08)	Shenango: Achieving High CPU Efficiency for Latency-sensitive Datacenter Workloads GrandSLAm: Guaranteeing SLAs for Jobs in Microservices Execution Frameworks Shinjuku: Preemptive Scheduling for µsecond-scale Tail Latency (optional)
Microservices + Hardware	Thursday (09/10)	µManycore: A Cloud-Native CPU for Tail at Scale Enhancing Server Efficiency in the Face of Killer Microseconds
	Tuesday (09/15)	HardHarvest: Hardware-Supported Core Harvesting for Microservices SIMR: Single Instruction Multiple Request Processing for Energy-Efficient Data Center Microservices ALTOCUMULUS: Scalable Scheduling for Nanosecond-Scale RPCs (optional)
09/15: Deadline for Project Proposal (by 9:00am)
Serverless Computing + Software Characterization and Virtualization	Thursday (09/17)	Serverless in the Wild: Characterizing and Optimizing the Serverless Workload at a Large Cloud Provider XFaaS: Hyperscale and Low Cost Serverless Functions at Meta
	Tuesday (09/22)	Firecracker: Lightweight Virtualization for Serverless Applications Catalyzer: Sub-millisecond Startup for Serverless Computing with Initialization-less Booting
Serverless Computing + Software Optimizations	Thursday (09/24)	MXFaaS: Resource Sharing in Serverless Environments for Parallelism and Efficiency SpecFaaS: Accelerating Serverless Applications with Speculative Function Execution Reflections and Optimizations for Cold Start Latency in Production Serverless Systems (optional)
	Tuesday (09/29)	Faa$T: A Transparent Auto-Scaling Cache for Serverless Applications Concord: Rethinking Distributed Coherence for Software Caches in Serverless Environments Fault-tolerant and transactional stateful serverless workflows (optional)
Serverless Computing + Hardware	Thursday (10/01)	Architectural Implications of Function-as-a-Service Computing Mosaic: Harnessing the Micro-Architectural Resources of Servers in Serverless Environments
	Tuesday (10/06)	BabelFish: Fusing Address Translations for Containers Sabre: Hardware-Accelerated Snapshot Compression for Serverless MicroVMs Serverless Computing on Heterogeneous Computers (optional)
Datacenter Tax Accelerators	Thursday (10/08)	A Hardware Accelerator for Protocol Buffers Dagger: Efficient and Fast RPCs in Cloud Microservices with Near-Memory Reconfigurable NICs CDPU: Co-designing Compression and Decompression Processing Units for Hyperscale Systems (optional)
	Tuesday (10/13)	Intel Accelerators Ecosystem: An SoC-Oriented Perspective : Industry Product AccelFlow: Orchestrating an On-Package Ensemble of Fine Grained Accelerators for Microservice Workloads
Midterm Project Presentations	Thursday (10/15)	Presentation Session 1
	Tuesday (10/20)	Presentation Session 2
10/22: Deadline for Midterm Report (by 9:00am)
Large Code Footprints of Datacenter Workloads	Thursday (10/22)	AsmDB: Understanding and Mitigating Front-End Stalls in Warehouse-Scale Computers Propeller: A Profile Guided, Relinking Optimizer for Warehouse-Scale Applications BOLT: A Practical Binary Optimizer for Data Centers and Beyond (optional)
	Tuesday (10/27)	Whisper: Profile-Guided Branch Misprediction Elimination for Data Center Applications I-SPY: Context-Driven Conditional Instruction Prefetching with Coalescing PDIP: Priority Directed Instruction Prefetching (optional)
AI Workloads in Datacenters	Thursday (10/29)	In-Datacenter Performance Analysis of a Tensor Processing Unit The Architectural Implications of Facebook's DNN-Based Personalized Recommendation MTIA: First Generation Silicon Targeting Meta's Recommendation Systems (optional)
	Tuesday (11/03)	Orca: A Distributed Serving System for Transformer-Based Generative Models Splitwise: Efficient Generative LLM Inference Using Phase Splitting Pre-Gated MoE: An Algorithm-System Co-Design for Fast and Scalable Mixture-of-Expert Inference (optional)
ML for Cloud and Datacenter Systems	Thursday (11/05)	Sinan: ML-based and QoS-aware Resource Management for Cloud Microservices SelfTune: Tuning Cluster Managers Twig: Multi-Agent Task Management for Colocated Latency-Critical Cloud Services (optional) FIRM: An Intelligent Fine-grained Resource Management Framework for SLO-Oriented Microservices (optional)
	Tuesday (11/10)	Learning Memory Access Patterns ECO: An LLM-Driven Efficient Code Optimizer for Warehouse Scale Computers Learning to Walk: Architecting Learned Virtual Memory Translation (optional)
Datacenter Architecture with Power Management (Part 1)	Thursday (11/12)	Flex: High-Availability Datacenters with Zero Reserved Power Dynamo: Facebook's Data Center-Wide Power Management System
	Tuesday (11/17)	SmartOClock: Workload- and Risk-Aware Overclocking in the Cloud SmoothOperator: Reducing Power Fragmentation and Improving Power Utilization in Large-scale Datacenters Thunderbolt: Throughput-Optimized, Quality-of-Service-Aware Power Capping at Scale (optional)
	Thursday (11/19)	Reserved as course buffer
Thanksgiving Fall Break
Datacenter Architecture with Power Management (Part 2)
	Tuesday (12/01)	Towards Energy Proportionality for Large-Scale Latency-Critical Workloads EcoFaaS: Rethinking the Design of Serverless Environments for Energy Efficiency ReTail: Opting for Learning Simplicity to Enable QoS-Aware Power Management in the Cloud (optional)
	Thursday (12/03)	TAPAS: Thermal- and Power-Aware Scheduling for LLM Inference in Cloud Platforms DynamoLLM: Designing LLM Inference Clusters for Performance and Energy Efficiency Rearchitecting Datacenter Lifecycle for AI: A TCO-Driven Framework (optional)
Final Project Presentations (Report due 12/15 by 9:00am)