Benoit Tellier created JAMES-3883:
-------------------------------------
Summary: About MimeTokenStream usage of LinkedList
Key: JAMES-3883
URL: https://issues.apache.org/jira/browse/JAMES-3883
Project: James Server
Issue Type: Improvement
Reporter: Benoit Tellier
A defacto programming MOJO around is that linked list is bad: overhead of
allocating bi-directional pointers, bad cache locality, etc... The benefits are
low compared to precised arrays that are a better de-facto choice.
Now see attached a flame graph from James showing 2% of its allocation, and
18.7 of memory allocations of that parsing use case. See attached flame graph.
I benchmarked ArrayDeque usage locally and it was shlightly faster (in the ms
range) and allocated ~1KB less per message:
{code:java}
BEFORE
JMHLongMultipartReadBench.benchmark1
avgt 5 39.842 ± 1.820 us/op
JMHLongMultipartReadBench.benchmark1:·gc.alloc.rate.norm
avgt 5 19448.003 ± 0.001 B/op
AFTER
JMHLongMultipartReadBench.benchmark1 avgt
5 38.687 ± 2.935 us/op
JMHLongMultipartReadBench.benchmark1:·gc.alloc.rate.norm avgt
5 18128.003 ± 0.001 B/op
{code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: server-dev-unsubscr...@james.apache.org
For additional commands, e-mail: server-dev-h...@james.apache.org
