Reasons to Use Cache

The UrbanSim simulations run on top of a file-based cache. The initial reason for the cache was to improve performance, and has turned out to be very useful for many features:

• The cache is much faster to access than the MySQL database, at least, so using the cache dramatically speeds up our processing.
• In many cases, a simulation will require more data than can fit in RAM. The simulation cache provides a place to store computed or non-computed values so that we only need to keep in memory just enough data for the computation at hand.
• Lag variables use the baseyear cache and simulation cache to quickly get data from prior years, often without having to re-compute it.
• When preparing for a simulation, UrbanSim uses the data from lag tables (see 4.4.1) in the baseyear to create historical data in the baseyear cache for years before the base year. This allows the simulation to run without any distinction between historical and predicted data.
• Lag variables on gridcell data need gridcell information from prior years. UrbanSim can use the information in the development_event_history table to create gridcell data for before the baseyear by ``unrolling'' the development events from before the base year.
• When looking for a primary attribute for a given dataset, Opus looks first in the current year. If the data is not in that year's cache, Opus automatically looks backward thru prior years until it finds it. This allows Opus to only store to cache primary attributes in the years in which they are modified, while still getting the performance and memory improvements by using the cache.
• The information in the simulation cache includes all of the intermediate variable values produced for every year. This allows us to inspect inspect this data for diagnostic purposes.
• Our indicator framework mines the cache data to produces charts, tables and maps that help to diagnose runs as well as providing input to policy decisions.