|
|
|
|
|
|
by c789a123
2819 days ago
|
|
|
My comments ( I am experienced in using spark, so could be biased): 1. as the worker machine as 16 vcpu, and 122G ram, setting spark.executor.memory to 8GB seems too small, I would be interested to see how it works with 32GB setting per 8 cores. 2. CDH is not so updated with spark releases, new spark releases 2.3 can be used together with CDH hadoop for test. 3. in table 4, spark is the only system without fails, which confirms it is a very robust system. 5. It is a performance test, but is the result verified? |
|
|