Using Branch Prediction and Speculative Execution for Magnetospheric Data Assimilation

Details Using Branch Prediction and Speculative Execution for Magnetospheric Data Assimilation Using Branch Prediction and Speculative Execution for Magnetospheric Data Assimilation

Dec 2005

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2005agufmsm13a0334d&link_type=abstract

American Geophysical Union, Fall Meeting 2005, abstract #SM13A-0334

Statistics

Computation

2722 Forecasting (7924, 7964), 7924 Forecasting (2722)

Scientific paper

Recent advances in the development of integrated models of the Sun-Earth environment are placing increasing emphasis on data assimilation schemes that can maximize the intelligence extracted from our sparse sampling of upwind conditions. Standard Kalman Filter techniques, widely used in tropospheric weather modeling, require significantly better coverage than is available upwind. To maximize the input of sparse upwind and magnetospheric data, and to reduce the forecast lead time computational penalty, we proposed to use Branch Prediction and Speculative Execution (BPSE) for data assimilation (Doxas and Horton,~2002). Branch Prediction and Speculative Execution consists of making probabilistic estimates of current upstream conditions, and distributing among available machines a large number of simulations that assume each of the probabilistically estimated states as initial conditions. As the near-Earth space evolves and near-Earth satellite data are compared with the models, some of the speculatively executed simulations will be seen to diverge from the observations more than others. At that point the machines that were executing them will be reassigned to new lines of speculative simulation, resulting in a continuous ensemble of runs that are in the neighborhood of the measured values. The scheme is particularly suited to Space Weather since our upwind early warning sentries can provide only sparse sampling of the incoming solar wind, while the bulk of our monitors, which can provide significantly better coverage, are located close to Earth and provide much shorter lead times. By the time the data come in from the near-Earth monitors, the forecasts of the speculative simulations are already in hand, reducing the lead time computational penalty (the portion of the lead time devoted to advancing the model) to almost zero. The scheme is similar to Ensemble Kalman Filters but is less reliant on dense data coverage, allows numerical models easier adherence to conservation laws, and can be used with empirical models without modification. Doxas, I., and W. Horton, Using Branch Prediction and Speculative Execution to Forecast Space Weather, Geomagnetic Environment Modeling conference, Telluride, CO (2002).

Affiliated with

Also associated with

No associations

Rating

Using Branch Prediction and Speculative Execution for Magnetospheric Data Assimilation does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Using Branch Prediction and Speculative Execution for Magnetospheric Data Assimilation, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Using Branch Prediction and Speculative Execution for Magnetospheric Data Assimilation will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-760126