Olivier conducted some tests for a Customer to find out how much performance impact SSL brings to the cluster. His conclusion was that the performance impact is quiet high.
[Link to internal document has been removed]
Here is a summary of the tests:
- Evaluation of performance impact by enabling SSL on Hadoop
- Setup: 4 Datanodes, 1 Namenode, Ambari 1.7, Hadoop 2.6, RedHat 6, RAM 128GB, 8 cores
- All network traffic was encrypted (RPC connections, WebInterfaces, MapReduce shuffe,...)
- Main tests: Copying files from and to HDFS (different file sizes 1-10GB) as well as running Teragen tests; Tests were conducted multiple times
- Performance impact when using RC4 is lower, however still not negligible
- Sample Result (5 GB file): no SSL ca. 20sec, AES 256bit RC4 ca. 100sec, AES 128bit/256bit 3DES ca. 800sec
- Wire encryption can cause high performance impact
Turning SSL and TDE on will definitely have a performance impact, as to how much in numbers, I don't know at the moment.