Add JMH benchmarks for OpenJDK

[nmittler]

No description provided.

[nmittler]

@normanmaurer @louiscryan @kruton @davidben Hey guys, I'd like to get started on benchmarking conscrypt. Any thoughts on what would be interesting benchmarks? Or if there is something I can lift from another project?

[nmittler]

One benchmark that I think would make sense is a side-by-side comparison between JDK, Netty, and Conscrypt. That will mean that it's just raw engine performance without ALPN/NPN ... just plain TLS.

[normanmaurer]

I think for example calling wrap(...) / unwrap(...) with different buffer sizes (src and dst) and also with heap / direct buffers and see how perf compares.

[nmittler]

@normanmaurer Yup good idea. Any thoughts on buffer sizes that might be interesting. Maybe keep it simple: 2 or 3 different buffer sizes should be sufficient ... otherwise the benchmarks will take forever and nobody will run them :)

[louiscryan]

Could try

http://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/file/567e410935e5/test/compiler/7184394

with Conscrypt vs JSSE

There's also

http://cr.openjdk.java.net/~shade/scratch/SHA256Bench.java

and maybe worth contributing to

https://github.com/ecki/jmhbench-cryptospeed

also

https://github.com/apache/commons-crypto/blob/77af2da3d2cc871fbfc3c6a75d6848e45be33794/src/test/java/org/apache/commons/crypto/CryptoBenchmark.java

(hadn't heard of this project before, might be worth sending them an email to talk about alignment)

[nmittler]

I've closed this for now. We have a simple benchmark for the engine. We'll expand the benchmark suite as we move forward, but we can open new issues as we go.