Tracing-based diagnosis in OpenStack

Project logistics 

• Mentors: Raja Sambasivan, Joe Talerico, Ata Turk, Peter Portante
• Min-max team size: 3-5
• Expected project hours per week (per team member): 6-8 
• Will the project be open source? Yes

Preferred past experience

• Familiarity w/OpenStack (Very useful)
• Experience coding w/Python and JavaScript (Very useful)
• Some knowledge of existing workflow-based tracing infrastructures, such as Dapper or OpenZipkin (Nice to have, but not necessary).

Project Overview: 

Workflow-centric tracing preserves the flow of causally-related activity within and across the components of a distributed system.  The rich information it provides makes it a promising substrate on which to layer many operational tasks, such as diagnosis and billing.  As such, workflow-centric tracing is gaining widespread adoption in industry.  Examples include Google’s Dapper and OpenZipkin.

In this project, we will extend OSProfiler, an in-development tracing infrastructure for OpenStack, with sophisticated functionality similar (or exceeding) that provided by existing production tracing infrastructures. We will then explore the utility of novel tracing-based diagnosis tools, such as Spectroscope (http://bit.ly/mkBbth), for helping debug problems in OpenStack.  Depending on time constraints, we will also explore new analyses and visualizations that utilize tracing for other important operational tasks.

Some Technologies you will learn/use: 

• Workflow-centric tracing, which is becoming increasingly popular in many large companies.
• How to debug/diagnose complex distributed systems
• OpenStack
• Python
• JavaScript
• Cutting edge research that uses workflow-centric tracing to automate/simplify various operational tasks.

Important background & related work: 

• Principled workflow-centric tracing of distributed systems (SoCC’16).  This paper systematizes knowledge about the past 10 years of research on tracing.
• Dapper: a large-scale distributed systems tracing infrastructure (TR Dapper-2010-1).  This paper is about Google’s tracing infrastructure.  
• PivotTracing: Dynamic Causal Monitoring for Distributed Systems (SOSP’15).  Another interesting tracing infrastructure.  
• Diagnosing performance changes by comparing request flows (NSDI’11). An application of tracing to localize the cause of performance regressions.
• Visualizing request-flow comparison to aid performance diagnosis in distributed systems (InfoVis’13).  
• OSProfiler: The start of a tracing infrastructure for OpenStack that we will build on.