Software Performance Fall 2013This is a course in the Masters program of the Faculty of Informatics at the University of Lugano. This page provides public access to the course material – lecture slides, videos, assignments.
Combine Systems, Tools, Theory
We connect material from Systems area courses (such as PL, OS, and HW) with Tools and Theory related to performance measurement and experimentation.
Navigate Across System Layers
Performance is a property that is best studied from a holistic perspective. For example, memory performance straddles garbage collector, virtual memory subsystem, and cache.
Become an Experimenter
We look at systems as objects of study and explore the factors affecting their performance. We consider workloads, measurement contexts, performance metrics, and data analysis.
Teaching Style How you get most out of this course
Why should you take this class?
This class prepares you to engineer efficient software systems, to detect performance problems and improve the performance of existing systems, and to identify possible new research problems.
How do you learn this?
You learn by trying, finding your own insights. This means the assignments can take a variable amount of time! We want you to formulate your own variations of problems and solve them.
How is the course structured?
- Readings and videos to prepare before class
- Interactive lectures
- Practical assignments after class
- Course: Midterm and Final exams
- Lab: Project
Logistics When, where, what and how
|Lectures||Wednesdays, 10:30 - 12:15 and Fridays, 13:30 - 15:15 in SI-006|
|Labs||Fridays, 15:30 - 17:15 in SI-006|
|Prerequisites||Java programming language experience|
|Textbook||David Lilja, Measuring Computer Performance, Cambridge University Press|
|Further Resources||Check our References page.|
|SP Course Grade||Homework: 40%, Midterm: 25%, Final: 35%|
|SP Lab Grade||Homework: 30%, Project: 70%|
Team Instructor and teaching assistants
We are a professor and two PhD students who are members of the Software and Programmer Efficiency research group of the Faculty of Informatics. Besides teaching this course, we spend our days (and sometimes nights) on research in the areas of programming languages, systems, and software engineering.
Schedule A one-semester, 14-week course
We have three weekly lecture slots, two for the course (white), and one for the lab (blue). Here is the tentative schedule. We will update it as we progress through the course.
|Date||Topic||Preparation||In-Class Activities||Assignment out||Assignment due|
|Wed, Sep. 18||Introduction||Refresh your Java, C, and Intel Assembly language knowledge||Introduction, Informa Session. Lecture 1 Slides.||A01. Java & C||-|
|Fri, Sep. 20||PL Program Representation||Watch videos on program representation
(~45 minutes of videos, plus some breaks):
|Lecture 2 Slides.||A02. Simple Bytecode Metrics||A01. Java & C|
|Fri, Sep. 20||Building the Jikes RVM||Read the Care and Feeding section of the Jikes RVM User's Guide||Install and build the Jikes RVM||L01. Getting Started with the Jikes RVM|
|Wed, Sep. 25||PL Control-Flow Graph||Control-flow graphs and dot. Lecture 3 Slides.||A03. Control-Flow Graphs with ASM||A02. Simple Bytecode Metrics|
|Fri, Sep. 27||PL Loops||Finding loops. Lecture 4 Slides.||A04. Dominator Analysis||A03. Control-Flow Graphs with ASM|
|Fri, Sep. 27||The Jikes RVM and Adaptive Optimization||Watch introductory videos on Jikes RDB
(~40 minutes of videos, plus some breaks):
|Jikes RVM, AOS. Lab 2 Slides.||L02. Jikes RVM Performance||L01. Getting Started with the Jikes RVM|
|Wed, Oct. 2||PL Compiler Optimization||Compiler optimization. Lecture 5 Slides.||A05. Add Exception Edges to Control-Flow Graph||A04. Dominator Analysis|
|Fri, Oct. 4||PL Program Analysis, PL Exceptions||Program analysis. Lecture 6 Slides. Exception handling.||-||-|
|Fri, Oct. 4||Getting Comfortable with the Jikes RVM||Jikes RVM in Eclipse, Lab 3 Slides.||L03. Look and Tickle||L02. Jikes RVM Performance|
|Wed, Oct. 9||PL Class Hierarchy, PL Call Graph||Pacman project, Lecture 7 Slides.||A06. Call Graph with ASM||A05. Add Exception Edges to Control-Flow Graph|
|Fri, Oct. 11||PL Method Call Optimization||Read Fast Static Analysis of C++ Virtual Function Calls by Bacon and Sweeney (Section 1 of and most of Section 2) and Optimization of Object-Oriented Programs using Static Class Hierarchy Analysis (CHA) by Dean, Grove, and Chambers (Section 1 and the first part of Section 2)||Lecture 8 Slides. Feedback on A02.||-||-|
|Fri, Oct. 11||Getting Ready for the Project||Feedback on L02. Questionnaire about L03.||L04. Project kick-off||L03. Look and Tickle|
|Wed, Oct. 16||PL Memory Management, Part 1||Read An Introduction to Garbage Collection: Part I - The Real Costs of C++ Memory Management||Stack, heap, globals, explicit memory management, reference counting. Lecture 9 Slides.||A07. Memory Management||A06. Call Graph with ASM|
|Fri, Oct. 18||PL Memory Management, Part 2||Read An Introduction to Garbage Collection: Part II - A Look Under the Hood||Heap analysis, NEW, tracing GC. Lecture 10 Slides.||-||-|
|Fri, Oct. 18||Project starts. From now on, we'll only hold lab sessions on request.||L04. Project kick-off|
|Wed, Oct. 23||PL Memory Management, Part 3||Memory Leaks. Leak Detection. Lecture 11 Slides. Paper: Low-Overhead Memory Leak Detection Using Adaptive Statistical Profiling||Optional A04 Resubmission. Dominator Analysis||A07. Memory Management|
|Fri, Oct. 25||HW Cache||Memory hierarchy. Access latency. Direct-mapped vs. set-associative vs. fully-associative caches. Lecture 12 Slides.||A08. Cache Simulator||Optional A04 Resubmission. Dominator Analysis|
|Fri, Oct. 25||Prepare demo of seed prototype||Demo of seed prototypes, meetings with mentor|
|Wed, Oct. 30||OS Dynamic Linking||Read The Inside Story on Shared Libraries and Dynamic Loading and Optional Program Library HOWTO (Linux article but the concepts are the same)||Shared libraries, dlopen/dlsym, library interposition, tracing dynamic linker activity. Lecture 13 Slides.||-||-|
|Wed, Nov. 6||Tools PL DTrace||Read Dynamic Instrumentation of Production Systems (all 14 pages)||DTrace. Lecture 14 Slides.||A09. Dynamic Instrumentation with DTrace||A08. Cache Simulator|
|Fri, Nov. 8||OS HW Virtual Memory||Virtual memory management. From malloc to mmap. Lecture 15 Slides.||-||A09. Dynamic Instrumentation with DTrace|
|Fri, Nov. 8||Prepare demo of improved prototype||Demo of improved prototypes, meetings with mentor|
|Wed, Nov. 13||Midterm exam||-||-|
|Fri, Nov. 15||PL Dynamic Class Loading||Read Dynamic Class Loading in the Java Virtual Machine (everything except Section 4 "Maintaining Type-safe Linkage")||Lazy loading, type-safe linking, multiple namespaces, class unloading, user-defined class loaders (SP-Compiler Eclipse project). Lecture 16 Slides.||A10. Practice with Dynamic Class Loading in the JVM||-|
|Fri, Nov. 15|
|Wed, Nov. 20||PL Dynamic Binary Instrumentation||Java instrumentation agent. ASM class rewriting. Lecture 17 Slides.||A11. Dynamic Binary Instrumentation||A10. Practice with Dynamic Class Loading in the JVM|
|Fri, Nov. 22||Tools PL R||Optionally read Evaluating the design of the R language||Variables, functions, vectors, factors, lists, data frames. Lecture 18 Slides.||-||-|
|Fri, Nov. 22||Prepare demo of improved prototype||Demo of improved prototypes, meetings with mentor||-||-|
|Wed, Nov. 27||Tools Theory Descriptive Statistics & Visualization||Read Chapters 1 "Introduction", 2 "Metrics of performance" and 3 "Average performance and variability" of the textbook||Central tendency & dispersion, PDF, CDF, jitter plot, box plot, histogram, ECDF, quantile/quantile plot. Lecture 19 Slides.||A12. Array Size Analysis||A11. Dynamic Binary Instrumentation|
|Fri, Nov. 29||Tools PL ggplot2||Read the ggplot2 Quick Reference||Grammar of Graphics, ggplot2. Lecture 20 Slides.||-||-|
|Fri, Nov. 29||-||-|
|Wed, Dec. 4||Discussion of Midterm||-||A12. Array Size Analysis|
|Fri, Dec. 6||Theory Systematic Error||Omitted Variable Bias, Profiler Bias. Lecture 21 Slides.||-||-|
|Fri, Dec. 6||Prepare demo of improved prototype||Demo of improved prototypes, meetings with mentor||-||-|
|Wed, Dec. 11||Theory Random Error||Read Chapter 4 "Errors in experimental measurements" of the textbook||Accuracy and random error vs. precision and systematic error, confidence intervals. Lecture 22 Slides.||-||-|
|Fri, Dec. 13||HW Branch Prediction||Read A study of branch prediction strategies||Why? Static vs. Dynamic. Bimodal. Two-Level Adaptive. Branch Target Prediction: BTB, RAS. Lecture 23 Slides.||-||-|
|Fri, Dec. 13||Prepare demo of improved prototype||Demo of improved prototypes, meetings with mentor||L05. Project presentation||-|
|Wed, Dec. 18||PL Call Profiling||Read gprof: A call graph execution profiler||Call graph. Calling context tree. Profiling with gprof, DTrace, hprof. DYI Java stack sampling. Lecture 24 Slides.||-||-|
|Fri, Dec. 20||Theory PL Complexity||Read Algorithmic Profiling||Cost Function, Asymptotic Complexity, Algorithmic Essence, Algorithmic Profiling. Lecture 25 Slides.||-||-|
|Fri, Dec. 20||Prepare presentation and demo of final system||Presentation and demo of final system||L05. Project presentation|
Homework consists of both written and programming assignments. You are encouraged to collaborate on the assignments. We recommend that you work in groups of two. However, you must produce the complete final writing and implementation on your own, and you must fully understand the solution you submit. Moreover, in your submission you must acknowledge all your collaborators.
In general, no late assignments are accepted.
The handouts of each assignment includes specific submission instructions. Please follow the submission instructions exactly as written!
Where are the assignments?
All assignments are posted on the schedule.