Issue link: https://read.dmtmag.com/i/43993
Environmental GIS Figure 2. An example LarvaMap run shows Walleye pollock larvae hatching at the mouth of Shelikof Strait, Alaska, and being swept on currents along the Alaska Peninsula. The line shows the centroid of the larvae cloud for each day of the model run. hardware to power all system aspects. The time required to complete the model was relative to the number and speed of CPUs available. RAM use wasn't as important, but to efficiently com- pute a large-scale particle run, 512MB of free RAM was required. Because there was only one model runner, multiple users would be required to complete any exist- ing model run before moving on to the next. Three chal- lenges were identified after Phase 1 of the project: 1. Support for multiple, simultaneous users 2. Support for multiple, simultaneous model runs 3. Improvement in particle-model run time Cloud-based architecture solved all three of these challenges by providing the ability to scale out the application horizontally, providing computational resources on demand. LarvaMap was re-architected for a cloud-based infrastructure and redeployed on the Amazon cloud. Front-end clients and middleware Web services were redesigned to be "hot deployable" to multiple instances and fronted with Amazon load balancers. This allowed the client and Web services to support an almost unlimited number of simultaneous users. To support multiple simultaneous model runs, the sys- tem required multiple model runners, and this required a way to manage communication among the Web service and runners. Using the Amazon Simple Queue Service, a system was developed that allowed a model run request to be serviced by any available model runner. Cloud resources improved the computation time of the model runners significantly. Modern laptop and desktop computers typically have anywhere from one to four CPU cores, meaning they can simultaneously perform one to four calculations at the same time. The particle model makes heavy use of all available CPU resources on computer hardware. To the model runner, a computer with only one core can only run the transport calculations for one particle at a time. If a computer has four cores, the model runner can perform four transport calculations at the same time, reducing processing time by 75 percent. Amazon offers a wide variety of instance types, each Figure 3. Ocean currents drive LarvaMap's modeling of particle dispersion. 20 GEO W ORLD / SEPTEMBER 2O11 with dedicated amounts of RAM and CPU resources (aws.amazon.com/ec2/instance-types). Because the model runner was highly dependent on CPU, and not RAM, a "high-CPU" instance was chosen with eight CPU cores and 7GB memory. This allowed multiple model runners to run on the same instance (reducing costs) and gave the runners enough processing power to complete large-scale model runs relatively quickly.