#466: Improve performance of Pointer.dump for large dumps. #518
Conversation
… dumps. Added simle unit test for Pointer.dump.
… dumps. Added simle unit test for Pointer.dump.
| StringWriter sw = new StringWriter(TITLE.length() + 2 + size * 2 + (size / BYTES_PER_ROW * 4)); | ||
| PrintWriter out = new PrintWriter(sw); | ||
| out.println(TITLE); | ||
| // byte[] buf = getByteArray(offset, size); |
There was a problem hiding this comment.
Based on discussion I can remove the commented code here and two lines below.
There was a problem hiding this comment.
Not a blocker, just curious: Were you going to remove these commented line?
There was a problem hiding this comment.
That goes back to the question in my original comment of this pull request, wether it's better to copy the buffer into a byte[] or to directly access the buffer. I tend to think that accessing the buffer is better, so based on that the commented lines could be removed.
|
I think this should be merged. Want to rebase/squash/cleanup @ck2510? |
|
I merged the base branch and pushed the result again. Let me know if there's anything else I can do. |
|
Drop the pom changes from the PR and this is good to go. |
… dumps. Pull and merge.
|
POM changes are gone now. |
Motivation: While we are in the channelReadComplete(...) method we might trigger flush() multiple times. This is wasteful as we will most likely do multiple sendto syscalls. We can do a better job by consolidate these flushes to just one and so be able to replace X sendto syscalls with 1 sendmmsg syscall. Modifications: - Keep track of if we are currently in the channelReadComplete(...) method and if so try to consolidate flushes. Result: Less syscalls
Also added simple unit test for Pointer.dump.
Initial implementation needed O(n) string allocations which was observed to be slow for large dumps. Using StringBuilder/StringWriter would bring down allocations to O(log n). With the implemented initial size estimate allocations should be down to O(1).
Also I tried to avoid copying the memory contents into an explicit byte[] buffer instead I'm getting the single bytes. Here it depends on the overhead of getByte() wether this is better or not. What is your opinion on that?