Issue link: https://read.dmtmag.com/i/43993
Where's the Output? Although leveraging the cloud provided solutions, it also introduced some complications. When the model runner was running on a single machine, it was pos- sible to write the output to the same machine. A Web server running on the same machine served up model results to the client. The bookkeeping was minimal. When splitting the model runner over many machines, however, an interesting problem was introduced: "How does the client know where to look for the model output?" Fortunately, this can be solved in a number of ways. The output can be saved to the machine on which it was run, and information can be stored in a database or simple lookup file. All requests could go through the REST service, which could locate the appropri- ate model output that was requested. This solution worked, but LarvaMap users wanted to bring model runners up and down with no effect. If the system started a model runner to handle a spike in traffic, but then was taken offline, the output also would be unavailable—this wasn't an option. The system needed to separate the computational instances from the model results—there needed to be a single location of model output storage. To accom- plish this, a distributed file system called GlusterFS (www.gluster.org) was implemented, running in the Amazon cloud. GlusterFS allows the system to have storage space in the cloud that's mountable and writable from any instance. The LarvaMap team mounted a 50GB GlusterFS storage system on each model runner and wrote output to it as a common location. A Web server was set up on a separate instance that connected to the model results and served the output. In addition, external applications in the cloud could utilize and serve the model output, such as ArcGIS server and THREDDS (www.unidata.ucar.edu/projects/THREDDS). In the future, LarvaMap implementers hope to expand the larval fish-inputs library to include data on more coastal and offshore spawning species as well as "generic" larvae options. Such species profiles would be based on agency data as well as published literature. In addition to the species library, the team hopes to incorporate biological behaviors such as the age at which larvae are strong enough to swim and their response to light and temperature. Currently, the model only can account for horizontal passive trans- port of particles, which isn't optimal for most larval fish species and/or older larval life stages, because they can swim against currents and move to or away from lit areas. Figure 4. LarvaMap's architecture provides common storage for all model output. Kyle Wilcox is a software engineer, Applied Science Associates; e-mail: KWilcox@asascience.com. Tiffany C. Vance is a geographer/IT specialist, Resource Assessment and Conservation Engineering Division, Alaska Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Atmospheric Administration; e-mail: Tiffany.C.Vance@noaa.gov. ADVERTISERINDEX ADVERTISER BAE Systems www.baesystems.com/gxp GEOINT www.geoint2011.com Esri www.esri.com/geoworldmob GeoWorld Reprints www.geoplace.com GeoWorld Subscriptions www.geoplace.com/subscribe USGIF www.usgif.org/membership 31 SEPTEMBER 2O11 / WWW . GEOPLA CE . COM 21 9 13 29 5 PAGE 32