While developing the Deep Space Trajectory Explorer, the team discovered that ordinary workstations didn’t have the computational power they needed. Artificially restricting the problem, or the space of potential solutions wasn’t an option for producing the best results, so the team was stuck waiting a long time for computations to complete.
“Even though each computation might take 10 milliseconds or less, when you need to perform millions of those computations, you’re going to be waiting – a full day in some cases. Constraining your problem by throwing away potential solutions results in losing important insights. But throwing cores and RAM at the problem means we don’t have to do that, and speeds our time to solution dramatically. With Oracle, we can spin up several bare metal compute instances, running our software, in minutes. The software then returns filtered and ranked results, which our engineers can then explore visually with our toolset,” said Sean M. Phillips, Principal Software Engineer at a.i. solutions.
“We don’t need that much hardware all the time, so it would never have been cost-effective to purchase it all. When we started exploring Oracle’s cloud, everything just worked – it was easy to get our JAVA 8 streaming API and Ubuntu environment up and running in minutes. Oracle Cloud Infrastructure, combined with third-party data streaming tools like Gluon’s CloudLink provided a strong technical implementation pattern to build on. We started in March, and by June we demonstrated the live solution running in real-time, displaying spacecraft trajectories to the audience, selecting and modifying them from a tablet.””
“Prior to using Oracle Cloud Infrastructure, the team was limited to computing about 25,000 independent trajectories per run – less than 2% of the typical design space. This forced analysts to spend a lot of time and energy attempting to organize the order of computations and estimate which areas of the design space were of highest priority, which risks missing potential solutions. Running the same task on Oracle Cloud Infrastructure Compute bare metal instances relieved these limitations. The numerical engine can now process, filter and return approximately 1.3 million independent trajectory maneuvers, a 52-fold improvement.”
“Someday, solutions like this have the potential to move beyond the design phase to the operational phase of missions,” Sean continued, “Imagine being able to instantly deploy a large cluster of servers to securely analyze every conceivable solution to a Mars mission trajectory problem, and select the best within minutes.”