wy96f commented on issue #2666: ARTEMIS-2336 Use zero copy to replicate 
journal/page/large message file
URL: https://github.com/apache/activemq-artemis/pull/2666#issuecomment-492539004
 
 
   @franz1981 @clebertsuconic  Thanks for the review. I've modified as you 
commented.  There is no issue in our production env using buffer send:) We just 
want to squeeze performance(sometimes tens of GB files needs to be transferred 
although it's only happening during catchup). Also in a few months, we'll use 
10-gigabit cards for our cluster which maybe make cathup more fast by using 
sendfile.  We deployed the test and saw that the change saved a visible amount 
of CPU. 
   Flamegraph of master:
   
![Flamegraph-orig](https://user-images.githubusercontent.com/7719761/57756416-c7238580-7725-11e9-80e9-01f4ce120b9f.png)
   
   Flamegraph of the PR:
   
![Flamegraph-zero](https://user-images.githubusercontent.com/7719761/57756442-d60a3800-7725-11e9-99c0-27108a7e551e.png)
   
   As we can see, at least 29% of the total time(21% FileChannelImpl::read and 
8% PacketImpl::encode) is saved.

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