Tool Infrastructure
           
Event Type Start Time End Time Rm # Chair  

 

Paper 10:30AM 11:00AM 38-39 Allen Malony (University of Oregon)
 
Title:
MRNet: A Software-Based Multicast/Reduction Network for Scalable Tools
  Speakers/Presenter:
Philip C. Roth (University of Wisconsin-Madison), Dorian C. Arnold (University of Wisconsin-Madison), Barton P. Miller (University of Wisconsin-Madison)

 

Paper 11:00AM 11:30AM 38-39 Allen Malony (University of Oregon)
 
Title:
The Tool Daemon Protocol (TDP)
  Speakers/Presenter:
Barton Miller (University of Wisconsin), Ana Cortés (Autonomous University of Barcelona), Miquel Senar (Autonomous University of Barcelona), Miron Livny (University of Wisconsin)

 

Paper 11:30AM 12:00PM 38-39 Allen Malony (University of Oregon)
 
Title:
Conservative Scheduling: Using Predicted Variance to Improve Scheduling Decisions in Dynamic Environments
  Speakers/Presenter:
Lingyun Yang (Department of Computer Science, University of Chicago), Jennifer M. Schopf (Math&Computer Science Division, Argonne National Laboratory), Ian Foster (Math&Computer Science Division, Argonne National Laboratory)
             

 

     
  Session: Tool Infrastructure
  Title: MRNet: A Software-Based Multicast/Reduction Network for Scalable Tools
  Chair: Allen Malony (University of Oregon)
  Time: Tuesday, November 18, 10:30AM - 11:00AM
  Rm #: 38-39
  Speaker(s)/Author(s):  
  Philip C. Roth (University of Wisconsin-Madison), Dorian C. Arnold (University of Wisconsin-Madison), Barton P. Miller (University of Wisconsin-Madison)
   
  Description:
  We present MRNet, a software-based multicast/reduction network for building scalable performance and system administration tools. MRNet supports multiple simultaneous, asynchronous collective communication operations. MRNet is flexible, allowing tool builders to tailor its process network topology to suit their tool's requirements and the underlying system's capabilities. MRNet is extensible, allowing tool builders to incorporate custom data reductions to augment its collection of built-in reductions. We evaluated MRNet in a simple test tool and also integrated into an existing, real-world performance tool with up to 512 tool back-ends. In the test tool, MRNet's performance was comparable to that of previous tool infrastructure. In the real-world tool, we used MRNet not only for multicast and simple data reductions but also with custom histogram and clock skew detection reductions. Although the tool's start-up protocol was already highly tuned, our tests of MRNet with 512 tool back-ends show significant improvements in average start-up latency.

This paper has been nominated for the Best Student Paper of SC2003 award.
  Link: Download PDF
   

 

     
  Session: Tool Infrastructure
  Title: The Tool Daemon Protocol (TDP)
  Chair: Allen Malony (University of Oregon)
  Time: Tuesday, November 18, 11:00AM - 11:30AM
  Rm #: 38-39
  Speaker(s)/Author(s):  
  Barton Miller (University of Wisconsin), Ana Cortés (Autonomous University of Barcelona), Miquel Senar (Autonomous University of Barcelona), Miron Livny (University of Wisconsin)
   
  Description:
  Runtime tools are crucial to program development. In desktop environments, we take tools for granted. In the Grid, it is difficult to find tools because of the complex interactions between applications, operating system and layers of job scheduling/management software. Therefore each runtime tool must be ported to run under each job management system; for m tools and n environments, the problem becomes an m*n effort, rather than m+n. The consequence is a paucity of tools in distributed and Grid computing environments.

In response, we analyzed several scheduling environments and runtime tools to better understand their interactions. We isolated what we believe are the essential interactions between tools, schedulers, resource manager, and applications. We propose a new standard, called the Tool Daemon Protocol, that codifies these interactions and provides the necessary communication functions. We implemented a pilot library and experimented with Parador, a prototype using the Paradyn Performance tools under Condor.
  Link: Download PDF
   

 

     
  Session: Tool Infrastructure
  Title: Conservative Scheduling: Using Predicted Variance to Improve Scheduling Decisions in Dynamic Environments
  Chair: Allen Malony (University of Oregon)
  Time: Tuesday, November 18, 11:30AM - 12:00PM
  Rm #: 38-39
  Speaker(s)/Author(s):  
  Lingyun Yang (Department of Computer Science, University of Chicago), Jennifer M. Schopf (Math&Computer Science Division, Argonne National Laboratory), Ian Foster (Math&Computer Science Division, Argonne National Laboratory)
   
  Description:
  In heterogeneous and dynamic environments, efficient execution of parallel computations can require mappings of tasks to processors with performance that is both irregular and time-varying. While adaptive domain decomposition techniques have been used to address heterogeneous resource capabilities, temporal variations in those capabilities have seldom been considered. We propose a conservative scheduling policy that uses information about expected future variance in resource capabilities to produce more efficient data mapping decisions. We first present techniques for predicting CPUload at some future time point, average CPUload for some future time interval, and variation of CPUload over some future time interval. We then present a family of stochastic scheduling algorithms that exploit such predictions when making data mapping decisions. Finally, we describe experiments in which we apply our techniques to an astrophysics application. The results demonstrate that conservative scheduling can produce execution times that are significantly faster and less variable than other techniques.
  Link: Download PDF