RFR: 8318756 Create better internal buffer for AEADs

Anthony Scarpino ascarpino at openjdk.org
Tue Nov 28 20:33:10 UTC 2023


On Thu, 23 Nov 2023 12:13:45 GMT, Daniel Jeliński <djelinski at openjdk.org> wrote:

>> Hi,
>> 
>> I need a review for a new internal buffer class called AEADBufferStream.  AEADBufferStream extends ByteArrayOutputStream, but eliminates some data checking and copying that are not necessary for what GaloisCounterMode.java and ChaCha20Cipher.java need.  
>> 
>> The changes greatest benefit is with decryption operations.  ChaCha20-Poly1305 had larger performance gains by adopting similar techniques that AES/GCM already uses. 
>> 
>> The new buffer shows up to 21% bytes/sec performance increase for decryption for ChaCha20-Poly1305 and 12% for AES/GCM.  16K data sizes saw a memory usage reduction of 46% with and 83% with ChaCha20-Poly1305.  These results come from the JMH tests updated in this request and memory usage using the JMH gc profile gc.alloc.rate.norm entry
>> 
>> thanks
>> 
>> Tony
>
> test/micro/org/openjdk/bench/javax/crypto/full/BenchBase.java line 122:
> 
>> 120:     public void decrypt() throws Exception {
>> 121:         decryptCipher.init(Cipher.DECRYPT_MODE, ks,
>> 122:             encryptCipher.getParameters().
> 
> Consider using a different method to obtain parameter spec here; the flamegraph suggests that `getParameters` is responsible for up to 20% of the time spent in these benchmarks

Thanks for point that out.  I have decided to leave this as is.  There are optimizations that can be made to `getParameters` that I would like the benchmark to measure.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/16487#discussion_r1408383879


More information about the security-dev mailing list