Changes between Version 12 and Version 13 of WikiStart


Ignore:
Timestamp:
04/12/14 16:29:31 (8 years ago)
Author:
jhammond
Comment:

--

Legend:

Unmodified
Added
Removed
Modified

  • WikiStart

    v12 v13  
    1 # Summary 
    2  
    3 OSPRI is a new one-sided runtime communication system for leadership-class supercomputers.  It aims to provide more flexible semantics compared to existing one-sided communication systems (e.g. [http://www.emsl.pnl.gov/docs/parsoft/armci/ ARMCI]) and target modern architectural trends by providing a thread-safe implementation. 
    4  
    5 Unlike its predecessor, [https://trac.mcs.anl.gov/projects/a1/ A1], OSPRI will provide only functionality above and beyond what is available MPI, hence is designed to run side-by-side with MPI, not as an independent runtime. 
    6  
    7 The proper citation of OSPRI is: 
    8  
    9 OSPRI: J. R. Hammond, J. Dinan, P. Balaji, I. Kabadshow, S. Potluri, and V. Tipparaju, ''The 6th Conference on Partitioned Global Address Space Programming Models'' (PGAS). Santa Barbara, CA, October 2012. "OSPRI: An Optimized One-Sided Communication Runtime for Leadership-Class Machines." 
    10  
    11 The corresponding Bibtex is: 
    12  
    13 {{{ 
    14 @inproceedings{Hammond:PGAS:2012:OSPRI, 
    15    author = {Jeff R. Hammond and James Dinan and Pavan Balaji and Ivo Kabadshow and Sreeram Potluri and Vinod Tipparaju}, 
    16    title = {{OSPRI}: An Optimized One-Sided Communication Runtime for Leadership-Class Machines}, 
    17    booktitle = {The 6th Conference on Partitioned Global Address Space Programming Models ({PGAS})}, 
    18    month = oct, 
    19    year = {2012}, 
    20    location = {Santa Barbara, CA, USA}, 
    21    url = {http://www.mcs.anl.gov/publications/paper_detail.php?id=1545}, 
    22 } 
    23 }}} 
    24  
    25 Please see the [preprint](http://www.mcs.anl.gov/publication/ospri-optimized-one-sided-communication-runtime-leadership-class-machines) for details. 
    26  
    27 # Goals of the project 
    28  
    29 * maximum communication performance 
    30 * scalability to millions of processes 
    31 * thread-safety 
    32 * support for hybrid systems (e.g. GPUs) 
    33 * modular design ala MPI 
    34 * software quality (readable, documented source code and robust build system) 
    35 * productive interaction with vendors 
    36  
    37 # Developers 
    38  
    39 Jeff Hammond, Sayan Ghosh, Sreeram Potluri, and Pavan Balaji have all contributed to this project. 
     1Moved to https://github.com/jeffhammond/ospri