Spatial Data Management Challenges in the Simulation Sciences Thomas Heinis, Farhan Tauheed, Anastasia Ailamaki
Simulation-Based Scientific Discovery Simulating natural phenomena leads to in-depth understanding!
Simulation Background Multitude of Analysis & Update Queries SIMULATION Analysis Queries MONITOR Time step time MONITOR SIMULATION Time step 1 Time step 2 Time step 3 → Massive Updates, Few Queries → In-Memory Spatial Indexing
In-Memory Spatial Indexing Using current approaches: Time breakdown: Most potential in reducing comparisons!
Research Directions Avoid data-oriented partitioning Avoid hierarchical structure → use uniform space-oriented partitioning → Configuration of the grid is difficult! Compression has only limited potential! Cache alignment is crucial
Massive Changes Simulations cause massive changes E.g., neural plasticity simulation: all elements move average change of 0.04μm in 285μm3 only 0.5% move more than 0.1μm Considerable indexing overhead State-of-the-art: moving object indexes assume known trajectory grace windows shift overhead to query execution 38%
Research Directions Use data itself! Use uniform grid as index Inherent connectivity Added connectivity (FLAT [ICDE 2012]) Use uniform grid as index Small overhead to build Only few objects change cell → few updates required → Tunable parameter grid resolution Big cells → few updates but inefficient queries Small cells → many updates but efficient queries Simulation Timestep
Spatial Indexing for the Simulation Sciences Challenges are not independent New class of spatial indexes: Efficient in main memory Efficient support of massive updates Likely based on grid: quick to build & only few updates needed More challenges ahead! New design point: probably not going to be as fast as known spatial indexes, but there is the trade off of updating fast and querying Out of core model building (large-scale spatial join) In-situ compression New storage media
Spatial Data Management Challenges in the Simulation Sciences Thomas Heinis, Farhan Tauheed, Anastasia Ailamaki Thank You!