pnowojski commented on issue #10529: [FLINK-15171] [serialization] fix 
performance regression caused by too many buffer allocations on string 
serialization
URL: https://github.com/apache/flink/pull/10529#issuecomment-566652560
 
 
   Thanks for the update @shuttie I will be waiting for your next update.
   
   > SerializationFrameworkMiniBenchmarks uses HybridMemorySegment, which is 
off-heap by default.
   
   Indeed that could be causing some differences.
   
   Regarding the `char[]` allocations, I was basing on those data:
   
   ```
        bytes  percent  samples  top
     ----------  -------  -------  ---
     8222540128   32.45%    40779  byte[]
     7509810768   29.64%    37258  char[]
     4320201040   17.05%    21491  java.lang.String
   ```
   
   ```
        bytes  percent  samples  top
     ----------  -------  -------  ---
     7591122240   29.43%     3271  char[]
     5360582240   20.78%     2243  java.lang.ThreadLocal$ThreadLocalMap
     5147640184   19.96%     2231  java.lang.String
   ```
   As I wrote in the previous message, if you compare relative `char[]` vs 
`String` allocations, it seems more or less fine. Previously I was looking just 
at the numbers `3271` vs `37258`, which didn't look fine.
   
   Re the `ThreadLocalMap` overhead, we can even see it in the number of 
allocations. Maybe that's what causing the GC pressure?

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to