maskit commented on pull request #7253: URL: https://github.com/apache/trafficserver/pull/7253#issuecomment-712476430
I guess, to save memory (freelist)? Using a bigger block size probably make sense if average packet size is going to be bigger. In fact, I tried changing the numbers (2K x 32) and it worked fine. Maybe we can make the change on another PR. However, the real issue here is not the number of blocks actually. The cost for freeing IOBufferBlocks doesn't seem ignorable, and the function has to do the free on ET_UDP if it has even one IOBufferBlock. With this change, the free only happens on ET_NET (I checked only QUIC use case). I found the issue on a debug build while I'm benchmarking H3, and the free was the most heaviest function. The issue disappears on non-debug builds somehow, but I think it's a nice to have on non-debug build as well. ---------------------------------------------------------------- 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]
