I've recently started working in the big data space and it definitely seems like the primary use case for Hadoop these days is HDFS. Spark seems to have entirely subsumed Hadoop MapReduce for most batch processing workloads.