You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@arrow.apache.org by "David Li (Jira)" <ji...@apache.org> on 2022/01/14 22:01:00 UTC
[jira] [Commented] (ARROW-15340) [C++] Try and fetch IPC footer in one read instead of two
[ https://issues.apache.org/jira/browse/ARROW-15340?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17476448#comment-17476448 ]
David Li commented on ARROW-15340:
----------------------------------
And just for reference, the Parquet reader already implements this optimization (though AIUI it's more helpful since Parquet footers are larger)
> [C++] Try and fetch IPC footer in one read instead of two
> ---------------------------------------------------------
>
> Key: ARROW-15340
> URL: https://issues.apache.org/jira/browse/ARROW-15340
> Project: Apache Arrow
> Issue Type: Improvement
> Components: C++
> Reporter: Weston Pace
> Priority: Major
>
> To concisely read the IPC footer we must do two reads. First read the footer length and then read the footer contents.
> On high latency filesystems we can cut down some of this startup time by assuming the footer is going to be relatively small and reading the last ~1MB of the file. Then we only need to do two reads if the footer is larger.
> This means most of the time we can usually get away with 1 read at the expense of possibly reading more data than needed.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)