Fiches de cours 2017-2018

PDF
 

Understanding Datacenter Software Dynamics

CS-719

Lecturer(s) :

Sites Richard Lee

Language:

English

Frequency

Only this year

Remarque

Next time: Fall 2017 - "Only this year"

Summary

Datacenter computing favors response time over throughput. This contrasts deeply with batch environments that prize throughput or total computation accomplished but ignore latency entirely. The focus of this course is understanding and controlling sources of latency, especially the long-tail delay.

Content

The focus of this course is understanding and controlling sources of latency, especially the long-tail delays that ruin a user experience. To do so, we will design and build software tools to measure not only the latency of various software and hardware subsystems, but to record the unexpected ways that these layers interact. It is precisely these unexpected dynamics that produce the most difficult latency problems, which almost all turn out to be easy to fix or control once the actual dynamics are visible. Controlling slow response times results in more efficient server use, which is worth literally millions of dollars within the datacenter industry. Your tools will run in a tiny datacenter environment and reveal real delays and real program interference.

Prerequisite skills: ability to program in C or C++; ability to edit files and run commands on Linux; familiarity with data structures and computer architecture at an undergraduate level; familiarity with multithreaded programs and asynchronous I/O; a dogged sense of curiosity.

 

Keywords

data centers
latency

Learning Prerequisites

Important concepts to start the course

concurrency, systems programming, operating systems, computer architecture

In the programs

Reference week

 
      Lecture
      Exercise, TP
      Project, other

legend

  • Autumn semester
  • Winter sessions
  • Spring semester
  • Summer sessions
  • Lecture in French
  • Lecture in English
  • Lecture in German