[
https://issues.apache.org/jira/browse/CALCITE-1091?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15151401#comment-15151401
]
Josh Elser commented on CALCITE-1091:
-------------------------------------
I've been looking into Avatica with a fine-tooth comb (profiler) the past week
or so. The children here are all issues that I've identified along the way and
have been testing with Phoenix.
The collective impact of all of them is really not that impactful, but Avatica
already wasn't *that* slow. Roughly speaking, I started out around with the
99th percentile of HTTP requests handled by the server ~250us. With these
changes, that numbers falls down to ~180us.
One thing I thought of today is that it would be a fun experiment to create a
NoopJdbcDriver specifically for the purpose of benchmarking Avatica without any
external impact (e.g. the actual database). It reminds of something
[YCSB|https://github.com/brianfrankcooper/YCSB]-esque, the ability to specify
some workload of mixed reads/writes and print some evaluation of the
performance for the high-level operation.
> Reduce Avatica RPC latency umbrella
> -----------------------------------
>
> Key: CALCITE-1091
> URL: https://issues.apache.org/jira/browse/CALCITE-1091
> Project: Calcite
> Issue Type: Bug
> Components: avatica
> Reporter: Josh Elser
> Assignee: Josh Elser
> Fix For: next
>
>
> I've been spending a bit of time looking at performance improvements to
> Avatica with the restriction of reducing the latency of operations (ignoring
> performance gains through concurrency).
> I'll use this as an umbrella for the issues I've run into so far.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
