Dylan Stark
On line
e'mail: dstark@cct.lsu.edu
twitter: dstark
TravelPod: dstark
The ParalleX Model
ParalleX is a new model of computation designed to overcome limitations of current parallel execution models by addressing the issues of latency, overhead, resource contention, and starvation which contribute to system performance degradation. The ParalleX strategy is to use multithreading, message-passing, global name space support, and lightweight synchronization elements to provide intrinsic system-wide latency hiding, near fine-grain global parallelism, and a unifed parallel programming model.
The target application domain for ParalleX is characterized by large, time-varying, and extremely sparse data structures, originating from informatics applications. These structures lack exploitable spatial and temporal locality, and are diffcult to load balance across a system. Furthermore, processing of these structures is data-directed and yields a low ratio of computation to communication. ParalleX addresses these issues through the use of multi-grain, decoupled transaction processing with asynchronous message-driven execution, supported by a distributed global address space, and dynamic load balancing.
The ParalleX project is being lead by Prof. Thomas Sterling, and is a part of the Computer Systems Science and Engineering focus area of the LSU Center for Computation & Technology.
HPC in the Clouds
An article in HPC Wire that looks at the applicability of cloud computing to the field of high performance computing, authored by Thomas Sterling and myself.
[article]
Preparing yerba mate
Yerba mate is a South American "tea". I made a video to demonstrate how to make a good mate.
[youtube, google video]
960 grid system
This is the CSS/XHTML template used for this site. It was created by Nathan Smith, and should be XHTML, CSS, and 508 compliant.
[homepage]