An iterative model that revisits the same civic graph many times runs far slower on classic disk-heavy MapReduce than on Spark. Why?
Select an answer to reveal the explanation.
Short Explanation
Classic MapReduce keeps slamming the disk between stages; Spark can keep working sets warmer in memory for those “again and again” civic graph loops. That is why iterative jobs often feel zippy on Spark. MapReduce is not secretly GPU-native for every hop.
Full Explanation
Spark is often faster than classic Hadoop MapReduce for iterative and in-memory workloads because it can cache working data and avoid writing every intermediate result to disk between steps. MapReduce does not default to keeping all intermediates in GPU memory; Spark does not forbid iteration; HDFS can store graph data independent of disabling Spark.