Add concurrency to Exponential Histogram

[lenin-jaganathan]

Guard the entire write operation externally for exponential histogram.

TODO:

• Add concurrency tests (I will try to come to this next week)

Fixes #5740

[lenin-jaganathan]

I initially thought that using a ReadWriteLock would provide a better performance than normal synchronization. But, after running the benchmarks, that is not the case. The overhead of ReadLock seems to be higher compared to the simpler math operations (like incrementing values) that are performed inside the lock. Considering the code complexity it brings it seems that not a good option. I am sharing benchmark results from my local PC (M1 10 cores),

Existing Implementation

1 - thread
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  96.115          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  85.981          ns/op

2 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  80.613          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  60.498          ns/op

4 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  211.090          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  179.947          ns/op

8 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  468.598          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  281.201          ns/op

Full Synchronization (the current state of PR)

1 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  96.945          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  85.520          ns/op

2 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  424.293          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  324.563          ns/op

4 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  922.629          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  853.906          ns/op

8 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  1832.407          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  1662.399          ns/op

Full Synchronization with AtomicLongArray (we have been using a AtomicLongArray until now)

1 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  94.078          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  84.147          ns/op

2 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  428.927          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  336.852          ns/op

4 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  865.197          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  817.838          ns/op

8 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  1708.773          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  1622.357          ns/op

ReadWrite Locks (Implementation in the first commit of this PR)

1 - thread
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  107.122          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2   90.638          ns/op

2 - threads
Benchmark                                                     
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  481.020          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  261.227          ns/op

4 - threads
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  946.514          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  888.778          ns/op

8 - threads                                                
CompareOTLPHistograms.oltpCumulativeExponentialHistogramTimer  avgt    2  4375.336          ns/op
CompareOTLPHistograms.oltpDeltaExponentialHistogramTimer       avgt    2  4036.524          ns/op

I will leave the ReadWriteLock implementation in the commit history so if someone has a look at it, then can run the tests as well (or correct me if I am doing something wrong there). I add a second commit to drop those changes and add synchronization to the writes.