Improving Workload Balance of a Marine CSEM Inversion Application

The marine Controlled-Source Electromagnetic Method (mCSEM) complements traditional seismic surveys for oil and gas exploration. A ship with a robust electromagnetic transmitter close to the seabed moves over an area previously equipped with fixed electromagnetic receivers on the seafloor. The colle...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2021 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW) s. 704 - 713
Hlavní autoři: Dagostini, Jessica Imlau, da Silva, Henrique Correa Pereira, Pinto, Vinicius Garcia, Velho, Roberto Machado, Gastal, Eduardo S. L., Schnorr, Lucas Mello
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.06.2021
Témata:
Conductivity
HPC
Load Balancing
mCSEM
MPI
Oil/Gas
Oils
Processor scheduling
Receivers
Scalability
Sea floor
Transmitters
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The marine Controlled-Source Electromagnetic Method (mCSEM) complements traditional seismic surveys for oil and gas exploration. A ship with a robust electromagnetic transmitter close to the seabed moves over an area previously equipped with fixed electromagnetic receivers on the seafloor. The collected data is then subject to data inversion techniques to compute the subsurface resistivity volume under the seabed. We characterize the workload imbalance as it originates from the scheduling policy and the mapping between measurement data and the resistivity model. We then propose a workload balancing improvement by applying the Sorted-Greedy scheduling policy in this data inversion application. Using realistic datasets, we demonstrate that the policy reduces the execution time from 32.1% up to 40.3% without affecting the results' numerical accuracy. Our changes also improved the application scalability, enabling the execution with a larger number of processes achieving additional gains from 53.1% up to 60.9%.
DOI:10.1109/IPDPSW52791.2021.00107