ByteBuffer allocation and de-allocation is expensive and becomes the performance bottleneck

[heyuanliu-intel]

During the performance benchmarking of the integration wildfly-openssl with Airlift, we found the performance bottleneck is ByteBuffer allocation and de-allocation.

Please refer to this project: https://github.com/heyuanliu-intel/AirliftPerformance

The system CPU utilization is about 20%. Most of the threads are blocked. Please refer to the thread dump file: https://raw.githubusercontent.com/heyuanliu-intel/AirliftPerformance/main/threaddump/thread1.txt

From this thread dump file, we can see that most of the threads are blocked by the ByteBuffer allocation and de-allocation.

After the investigation, we suggest to use pool based method to do ByteBuffer allocation and de-allocation.
The performance benchmarking is done.
Please refer to this project: https://github.com/heyuanliu-intel/ByteBufferPerformance

Based on the pool based allocation/de-allocation, we can get 4X performance improvement for the Airlift benchmarking.

I will submit the code and make the pull request.

[heyuanliu-intel]

Made a pull request #120.