Datacenter Simulation Methodologies

Full Day Tutorial: Sunday, Dec 14, 09:00-17:00 at MICRO 2014

In an era of big data, datacenters comprise the essential infrastructure for cloud computing. Yet simulation and evaluation methodologies remain a challenge as computer architects seek to improve datacenter performance and efficiency. This tutorial will demonstrate tools for datacenter research and walk participants through the approach taken at Duke. At the end of the tutorial, participants will be able to:

  • Deploy a full-system, cycle-accurate simulator.
  • Simulate datacenter workloads, such as web search and map reduce.
  • Evaluate heterogeneous processors, emerging memories within a server.
  • The tutorial will demonstrate the simulation of datacenter workloads using MARSSx86 and DRAMSim2. Moreover, we will survey queueing models and discrete event simulators for modeling datacenter dynamics. Finally, we will present case studies with this infrastructure.

    Tutorial Outline

    09:00-09:15       Introduction [slides]
    09:15-10:30       Setting up MARSSx86 and DRAMSim2 [slides]
    10:30-11:00       Break
    11:00-12:00       Spark simulation [slides]
    12:00-13:00       Lunch
    13:00-13:30       Spark continued
    13:30-14:30       GraphLab simulation [slides]
    14:30-15:00       Break
    15:00-16:15       WebSearch simulation [slides]
    16:15-17:00       Case studies [slides]

    Tutorial Organizers


    Prof. Benjamin C. Lee, Duke University (benjamin.c.lee@duke.edu)
    Qiuyun Wang, Duke University (qiuyun.wang@duke.edu)
    Seyed Majid Zahedi, Duke University (zahedi@cs.duke.edu)
    Tamara Silbergleit Lehman, Duke University (tamara.silbergleit@duke.edu)