3 ms·
You are thinking of Liquid Robotics[0]. I always found it interesting that they used a Java implementation of MapReduce to do their substantial onboard data pro
by jandrewrogers 3y ago
You are thinking of Liquid Robotics[0]. I always found it interesting that they used a Java implementation of MapReduce to do their substantial onboard data processing, which seemed very much at odds with the other issue of being extremely power constrained.
[0] https://www.liquid-robotics.com https://www.liquid-robotics.com
- fbdab103 3y agoIf you are doing all of the processing on a single node, isn't MapReduce just another name for batch processing?
- jandrewrogers 3y agoNot in this case. The onboard processing of data from the sensor suites was a Beowulf-style cluster of mobile phone ARM systems. All the computing hardware was carefully selected and engineered to be extremely frugal with power. Which is why I found it ironic that the software architecture was done in an overtly wasteful "throw hardware at it" style. Some basic attention to software performance engineering could have increased efficiency by an integer factor. It was quite the contradiction.