Hacker News Reader

The title should be changed. The article specifically ends with saying that Hadoop is not dead. Spark might have killed MapReduce, but the rest of Hadoop is alive and very useful. The only alternatives worth considering for replacing those parts of Hadoop are either proprietary or too dependant on a single company.

...and HiFrames [1] and HPAT [2] will kill Spark ;-)

HiFrames and HPAT are built on Julia [3] and use the Julia packages ParallelAccelerator.jl [4].

[1] https://arxiv.org/abs/1704.02341

[2] https://github.com/IntelLabs/HPAT.jl

[3] https://julialang.org/

[4] https://github.com/IntelLabs/ParallelAccelerator.jl

Spark still uses Hadoop underneath. While Hadoop mapreduce uses the disk, spark uses memory for faster processing. And I am not sure anything is going to kill spark soon. Once a library gains critical mass, it's harder to replace it in existing systems

Spark can read from and write to HDFS, and YARN (the resource negotiator in Hadoop v 2) is one of the cluster managers supported by Spark.

but you can run a Spark cluster without YARN--eg, with Mesos--or with the built-in manager provided in the Spark distro. Likewise, your Spark cluster doesn't need to read or write to HDFS.

Spark killed Hadoop