Deploying Condor Using VirtualBox

1
Deploying Condor Using VirtualBox on Windows
Hosts

• David Herzfeld1, Craig A. Struble, Ph.D2, and
  Lars E. Olson, Ph.D1
• Marquette University
• 1Department of Biomedical Engineering
• 2Department of Mathematics, Statistics, and
  Computer Science
• Southeast Wisconsin High Performance (SeWHiP)
  Computing Consortium (www.sewhip.org)

2
Introduction

• Most systems on Marquettes campus run Windows
  XP or Windows Vista
• Most researchers at Marquette require Linux based
  environments
• Most sys-admins have primarily Windows based
  experience

3
Project Goals

• Allow researchers to use underutilized Marquette
  systems
• Requires Linux environment
• Allow researchers full use of the Condor system
• check pointing, migration, etc.
• Do not change Windows users experience
• Easy installation and maintenance for Windows
  sys-admins

4
Possible Solutions

• Use Condor for Windows
• Very few researchers using Windows based
  applications
• Most require a GUI
• Windows Linux dual boot systems
• Management is difficult
• Windows users will be affected
• Condor Virtual Universe
• VMware
• Xen does not support Windows hosts
• Configuration Overhead
• Virtualization

5
Initial Solution CoLinux

• Based on previous work by Neeman, Serverini (U.
  of Oklahoma), and Sumanth (Nebraska-Lincoln),
  2006
• Runs natively in Windows
• Near native performance
• Small hardware dependencies
• Open source
• Relatively unsupported small community
• Only supports 32 bit environments
• No current SMP support
• Few recent developments

6
Current Solution Sun VirtualBox

• Free Open Source
• Capable or running under 32 64 bit Windows
• Guest Additions allow for near native speeds
• Some disk intensive operations may see a speed up
• Supported continually developed
• Large community support
• No current SMP support
• planned for future release

7
VirtualBox Windows
8
CondorMAC CoMon

• CondorMAC
• Allows for distribution via images
• Replaces the VirtualBox MAC address with a
  modification of the host MAC address
• CoMon
• Determine sec. since last key stroke
• Determine total processor utilization
• Determine VirtualBox processor utilization
• Write data to file in host-guest shared folder
• Visual Basic monitor service

9
hclust_service

• Fork headless VirtualBox machine at Windows
  startup.
• Remove any VirtualBox remnant machines
• Nicely shut down guest system when requested
  (via ACPI)

10
Condor Virtual Machine
11
Virtual Machine Basics

• Guest hostname is based on assigned IP address
• Central Manager name IP are determined
• Personalized text file in shared folder
• Variables replaced in condor_config.local
• For remote administration
• ssh key in shared folder
• Condor launches after guest startup
• init.d script

12
Benefits of Immutable Hard Drives

• Defines an upper limit on the amount of space
  Condor users can utilize (100 GB).
• Removal of the snapshots by hclust_service
  returns the VM to an original state.
• Returns unused hard drive space back to the
  Windows host.
• This space would not be returned if it were
  simply a dynamically expanding .vdi image

13
Condor Virtual Machine
14
VM Additions

• Additions included in immutable hard drive
• Software packages utilized by researchers
• Geant, OpenFOAM, AutoDock, etc.
• Sets up execution path for use in Condor jobs
• Package names and versions added to Condor
  ClassAds

15
Installation

• Distributed as a zip file, batch installer
• Installs VirtualBox
• Creates a new virtual machine
• MAC Address set by CondorMAC
• Memory computed as fraction of total system
  memory
• Bridged networking set up
• Adds CentOS image to virtual machine
• Sets services to run at system startup
• Starts the new VM
• Installation time is minutes

16
Virtualization Issues

• Networking Problems
• Current bridged connection requires an additional
  IP address for each worker
• Doubles the on campus IP addresses!

17
Future Work

• SMP Support
• Support for different networking modes
• using NAT, OpenVPN, etc.
• Specific VM updates via repositories
• Tailored distributions
• Bioinformatics
• Fluid Dynamics
• Neural Simulations

18
Acknowleldgments

• Information Technology Services Marquette
  University
• Kathy Lang CIO, Clifford Brown, Patrick Blume,
  Aaron Gember, Chad Gorectke
• Lingtao Zeng Marquette MSCS System Administrator
• Brad Bonczkiewicz Marquette Engineering System
  Administrator
• Biomedical Engineering Department
• Mathematics, Statistics and Computer Science
  Department

19
Questions/Comments?

• Thank you!