Scioto

From HPCRL Wiki
(Difference between revisions)
Jump to: navigation, search
(Download)
(Updated download link)
 
(47 intermediate revisions by 4 users not shown)
Line 1: Line 1:
== Scioto: Shared Collections of Task Objects ==
+
Scioto, or Shared Collections of Task Objects, is a lightweight framework for providing task management on
 +
distributed memory machines under one-sided and global-view parallel programming models. Scioto provides locality aware dynamic load balancing and interoperates with MPI, ARMCI, and Global Arrays.  Through task parallelism, the Scioto framework provides a solution for overcoming irregularity, load imbalance, and heterogeneity as well as dynamic mapping of computation onto emerging architectures.
  
Scioto (pronounced /sigh-OH-toe/ [http://en.wikipedia.org/wiki/Scioto_River]) is a framework for supporting task parallelism on distributed memory machines under one-sided and global view parallel programming models.  This framework allows the user to express their program as a set of tasks that are automatically scheduled and load balanced with respect to locality by the runtime system.    Through task parallelism, the Scioto framework provides an approach to overcoming irregularity, load imbalance, and heterogeneity as well as dynamic mapping of computation onto emerging architectures.
+
== Scioto Programming Model ==
  
Scioto uses ARMCI and MPI internallyWe have tested interoperability with MPI, ARMCI, and Global Arrays applications and are investigating interoperability with additional parallel programming tools.
+
Scioto provides the user with a task pool programming modelUnder this model the user expresses their computation as a set of ''tasks'' that can be executed in parallel.  Tasks read inputs from their task descriptor as well as from data stored in the global address space.  Tasks store their output in the global address space and are also permitted to add new tasks to the task collection.
  
== Acknowledgements ==
+
Below is pseudocode that shows the structure of a simple Scioto program.
  
Current project members:
+
<pre>
* James Dinan (Scioto maintainer) [dinan at cse dot ohio-state dot edu]
+
// The user decides how to store task arguments within the task
* D. Brian Larkins [larkins at cse dot ohio-state dot edu]
+
// body.  Generally, users define a struct to organize this data.
* Prof. P. Sadayappan [saday at cse dot ohio-state dot edu]
+
typedef struct {
* Sriram Krishnamoorthy [sriram at pnl dot gov]
+
  double args;
* Jaren Nieplocha [nieplocha at pnl dot gov]
+
} mytask_t;
 +
 
 +
// The user provides a function that implements the task.  This
 +
// function is run by Scioto when executing the task.
 +
void task_fcn(gtc_t gtc, task_t *my_task) {
 +
  // ...
 +
}
 +
 
 +
int main() {
 +
  gtc_t    tc  = gtc_create(...);      // Create a task collection
 +
  task_t  task = gtc_task_create(...); // Create a task descriptor
 +
  mytask_t my_t = gtc_task_body(task);  // Get the task body
 +
 
 +
  my_t.args = ...;    // Fill in the task's arguments
 +
 
 +
  gtc_task_add(task); // Add the task to the task collection
 +
 
 +
  gtc_process(tc);    // Enter task-parallel region
 +
}
 +
</pre>
 +
 
 +
== Documentation ==
 +
 
 +
=== Programming Interface ===
 +
 
 +
Documentation on Scioto's user API can be found here: [[Scioto API]]
 +
 
 +
=== Publications ===
 +
 
 +
'''Scalable Work Stealing'''
 +
[http://www.cse.ohio-state.edu/~dinan/research/tpool/dinan_sc09.pdf PDF]
 +
[http://www.cse.ohio-state.edu/~dinan/research/tpool/dinan_sc09_slides.ppt PPT]
 +
<br />
 +
James Dinan, Sriram Krishnamoorthy, D. Brian Larkins, Jarek Nieplocha, P. Sadayappan <br />
 +
Proc. 21st Intl. Conference on Supercomputing (SC). Portland, OR, Nov. 14-20, 2009.
 +
 
 +
'''Scioto: A Framework for Global-View Task Parallelism'''
 +
[http://www.cse.ohio-state.edu/~dinan/research/tpool/dinan_icpp08.pdf PDF]
 +
[http://www.cse.ohio-state.edu/~dinan/research/tpool/dinan_icpp08_slides.ppt PPT]
 +
<br />
 +
James Dinan, Sriram Krishnamoorthy, D. Brian Larkins, Jarek Nieplocha, P. Sadayappan <br />
 +
Proc. of 37th Intl. Conference on Parallel Processing. Portland, OR, Sept. 8-12, 2008.
 +
 
 +
== People ==
 +
 
 +
=== Project Leaders ===
 +
 
 +
* [http://www.mcs.anl.gov/~dinan James Dinan] (Primary contact)
 +
* [http://www.cse.ohio-state.edu/~saday Prof. P. Sadayappan]
 +
* [http://hpc.pnl.gov/people/sriram/ Sriram Krishnamoorthy]
 +
 
 +
=== Contributors ===
 +
 
 +
* [http://www.cse.ohio-state.edu/~larkins D. Brian Larkins]
 +
* [http://hpc.pnl.gov/people/jarek/ Jarek Nieplocha]
 +
 
 +
== Acknowledgements ==
  
This research was supported in part by DOE grant #DE-FC02- 06ER25755 and NSF
+
This research was supported in part by DOE grant ''#DE-FC02-06ER25755'' and NSF grant ''#0403342''.
grant #0403342.
+
  
 
== Download ==
 
== Download ==
  
We are currently providing preview development snapshots for download.
+
Please see the included README file for installation instructions.  If you have questions or comments, contact Jim Dinan (dinan at mcs.anl.gov).
  
Dev Snapshot, 09-10-2008 [http://www.cse.ohio-state.edu/~dinan/scioto/scioto-dev-09-10-2008.tar.gz]
+
* [http://hpcrl.cse.ohio-state.edu/scioto/scioto-r7688.tar.gz Scioto v0.2] -- ''11/15/2011''
 +
* [http://www.mcs.anl.gov/~dinan/download/scioto/scioto-0.1.tar.gz Scioto v0.1] -- ''11/15/2010''

Latest revision as of 16:25, 7 January 2015

Scioto, or Shared Collections of Task Objects, is a lightweight framework for providing task management on distributed memory machines under one-sided and global-view parallel programming models. Scioto provides locality aware dynamic load balancing and interoperates with MPI, ARMCI, and Global Arrays. Through task parallelism, the Scioto framework provides a solution for overcoming irregularity, load imbalance, and heterogeneity as well as dynamic mapping of computation onto emerging architectures.

Contents

Scioto Programming Model

Scioto provides the user with a task pool programming model. Under this model the user expresses their computation as a set of tasks that can be executed in parallel. Tasks read inputs from their task descriptor as well as from data stored in the global address space. Tasks store their output in the global address space and are also permitted to add new tasks to the task collection.

Below is pseudocode that shows the structure of a simple Scioto program.

 // The user decides how to store task arguments within the task
 // body.  Generally, users define a struct to organize this data.
 typedef struct {
   double args;
 } mytask_t;

 // The user provides a function that implements the task.  This
 // function is run by Scioto when executing the task.
 void task_fcn(gtc_t gtc, task_t *my_task) {
   // ...
 }

 int main() {
   gtc_t    tc   = gtc_create(...);      // Create a task collection
   task_t   task = gtc_task_create(...); // Create a task descriptor
   mytask_t my_t = gtc_task_body(task);  // Get the task body
   
   my_t.args = ...;    // Fill in the task's arguments

   gtc_task_add(task); // Add the task to the task collection

   gtc_process(tc);    // Enter task-parallel region
}

Documentation

Programming Interface

Documentation on Scioto's user API can be found here: Scioto API

Publications

Scalable Work Stealing PDF PPT
James Dinan, Sriram Krishnamoorthy, D. Brian Larkins, Jarek Nieplocha, P. Sadayappan
Proc. 21st Intl. Conference on Supercomputing (SC). Portland, OR, Nov. 14-20, 2009.

Scioto: A Framework for Global-View Task Parallelism PDF PPT
James Dinan, Sriram Krishnamoorthy, D. Brian Larkins, Jarek Nieplocha, P. Sadayappan
Proc. of 37th Intl. Conference on Parallel Processing. Portland, OR, Sept. 8-12, 2008.

People

Project Leaders

Contributors

Acknowledgements

This research was supported in part by DOE grant #DE-FC02-06ER25755 and NSF grant #0403342.

Download

Please see the included README file for installation instructions. If you have questions or comments, contact Jim Dinan (dinan at mcs.anl.gov).

Personal tools