UCoMS A Grid-enabled Workflow System for Reservoir Uncertainty Analysis Emrah Ceyhan, Gabrielle Allen, Chris White, Tevfik Kosar* Center for Computation & Technology Louisiana State University June 23, 2008 CLADE 08 AT LOUISIANA STATE UNIVERSITY
Roadmap UCoMS Project Reservoir Modeling & Uncertainty Analysis Computing and Data Challenges Data-Aware Workflow Solution Results and Conclusion
Oil Industry in Louisiana Major oil producing state in US: 5th in production 8th in reserves Home to 2 of 4 strategic petroleum reserves 17 petroleum refineries (capacity 2.8M barrels/ day) Ports receive ultra large oil tankers 20,000 oil producing wells, around 4K offshore.
UCoMS Ubiquitous Computing & Monitoring System for Discovery & Management of Energy Resources DOE/Louisiana BOR funded Petroleum engineering Wireless sensor networks Grid technologies Applications Reservoir simulation Uncertainty analysis, sensitivity studies, history matching Real-time well surveillance Drilling performance analysis with high-rate data
Reservoir Simulation Mathematical model for fluid flow in a reservoir involves density, permeability (K), mobility, pressure (P), production rate (q), porosity & saturation, where m denotes either oil, water or gas. Many geological parameters cannot be measured or modeled and are unknowns. We are using UTChem (3D, multiphase, multicomponent, compositional, variable temperature, FD simulator)
Reservoir Uncertainty Analysis Understand the effect of uncertainty in reservoir studies to guide development and operational decisions Uncertainty in different (geological) parameters (factors) pressure, permeability, water saturation, critical gas saturation, gas/water end points Factors (parameters) are classified into Controllable: Can be varied by process implementers, e.g. Well Location, injection rate, Observable: Can be relatively accurately measured but not controlled, e.g. Depth to a structure, Uncertain: Cannot be accurately measured or controlled, e.g. Permeability far from wells,
UCoMS Challenges Computation: Millions of simulations, each running 16-160 hours Data: Each simulation processing 40-400 MB of data More than 1PB data total Real-time flow of data from sensors to grid resources Sensor control and monitoring Automated allocation, location, transfer, and archiving \ Workflow: Coordination of Computation and Data
End-to-end Scenario Experiment Sensing & Control Sensor data Data Storage &Transfer EnKF Seismic Models UTChem/BlackOil Computing HPC & Grid resources 8
Data Evolution initial POROV =0.18 POROV =0.7 TRUTH 9
UCoMS Abstract Workflow
UCoMS Concrete Workflow
Workflow Expansion JOB i get JOB j put JOB k
Workflow Expansion JOB i JOB i Stage-in get JOB j Execute job j put JOB k Stage-out JOB k Compute Jobs Data Individual placement Jobs Jobs
Workflow Expansion JOB i JOB i Allocate space for input & output data JOB i Stage-in Stage-in get JOB j Execute job j Execute job j put JOB k Stage-out Release input space Stage-out Compute Jobs Data Individual placement Jobs Jobs JOB k JOB k Release output space
Separation of CPU & I/O DAG specification C Compute Job Queue DaP A A.data DaP B B.data Job C C.compute.. Parent A child B Parent B child C Parent C child D, E.. D A B C E Workflow Manager F E DaP Job Queue
Data-Aware Scheduler Type of a job? transfer, allocate, release, locate.. Priority, order? Protocol to use? Second vs Third party? Available storage space? Best concurrency level? Reasons for failure? Best network parameters? tcp buffer size I/O block size # of parallel streams
Data-Aware Scheduler Type of a job? transfer, allocate, release, locate.. Priority, order? Protocol to use? Second vs Third party? Available storage space? Best concurrency level? Reasons for failure? Best network parameters? tcp buffer size I/O block size # of parallel streams GridFTP
Separation of CPU & IO
Stork Transfer Methods regular: one connection per file, serial transfer multi-connection: one connection per file, concurrent transfer A A a small file a small file B B single-connection: one connection for all transfers A many small files B data-fusion: merge small files into larger chunks A a large file B 16
Stork Transfer Results
Monitoring DAGs via WEB
Monitoring DAGs via WEB
Monitoring DAGs via WEB
Summary Uncertainty Analysis in Reservoir Simulations can be both computationally and data intensive may require complex workflows We provide data-aware workflow management computation & I/O separated at lower levels data related tasks are handled by the data scheduler data optimizations made easier 21
This work has been sponsored by: DOE, NSF and LA BoR For more information UCoMS: http://www.ucoms.org Stork: http://www.storkproject.org PetaShare:http://www.petashare.org
Hmm.. This work has been sponsored by: DOE, NSF and LA BoR For more information UCoMS: http://www.ucoms.org Stork: http://www.storkproject.org PetaShare:http://www.petashare.org