Cloud Computing Architecture Project

This repository contains code, results and reports for the Cloud Computing Architecture course project at ETH Zurich. The project explores how to schedule latency-sensitive and batch applications in a cloud cluster.

• Part 1: Analyze the sensitivity of latency sensitive cloud application (memcache) with various types of resource interference (CPU, L1 instruction and data cache, L2 cache, L3 cache and Memory Bandwidth interference).
• Part 2: Analyze the sensitivity of batch cloud applications (PARSEC) to various types of resource interference (same as part 1) and parallelism.
• Part 3: Design an interference-aware scheduling policy to run latency-critical and batch applications on shared cluster.
• Part 4: Design an interference-aware scheduling policy to run the latency-critical applications on the same VM.

Results

• The static scheduler for part 3 achieves an SLO of 2ms for 95th percentile latency at 30k QPS.
• The dynamic scheduler for part 4 achieves an SLO of 2ms for 95th percentile latency for loads varying between 5k QPS and 100k QPS with 8 second intervals in between next waves.

Files and Folders

• cloud_comp_project_parts1-4.pdf: This file contains the detailed project description and instructions for setting up the clusters required for the project.
• interference: This folder contains the YAML files used induce resource contention in the clusters.
• memcache: This folder contains the YAML files used to set up memcache pod for parts 1 & 3.
• parsec-benchmarks: This folder contains the YAML files used to set up PARSEC pods for parts 2 & 3.
• project_yaml_files: This folder contains the YAML files we used to start up the clusters used in parts 1-4.
• reports: This folder contains our reports for the project. You can find the detailed explanations of memcache and PARSEC application profiling in report_part_1_2.pdf file. You can find detailed explanation of the design of our static and dynamic scheduling policies, and their respective performances in the report_part_3_4.pdf.
• results: This folder contains the output logs of memcache and PARSEC executions for parts 1-4.
• src: This folder contains the implementations of our scheduling policies. static_schedule.py contains the static scheduling policy for part 3. controller_final folder contains the implementation of the dynamic scheduling policy and controller for part 4.

Contributors

• Julia Bazinska
• Floris Westermann
• Berke Egeli