aurora-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Joshua Cohen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (AURORA-868) Decouple initial scheduling delay from first scheduling penalty
Date Sat, 18 Oct 2014 00:14:33 GMT

    [ https://issues.apache.org/jira/browse/AURORA-868?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14175738#comment-14175738
] 

Joshua Cohen commented on AURORA-868:
-------------------------------------

https://reviews.apache.org/r/26901/

> Decouple initial scheduling delay from first scheduling penalty
> ---------------------------------------------------------------
>
>                 Key: AURORA-868
>                 URL: https://issues.apache.org/jira/browse/AURORA-868
>             Project: Aurora
>          Issue Type: Task
>          Components: Scheduler
>            Reporter: Bill Farner
>            Assignee: Joshua Cohen
>
> When the scheduler inserts a task into the scheduling queue, it uses a {{BackoffStrategy}}
to calculate the initial penalty.  The initial value is tied to {{-initial_schedule_delay}}.
> Based on data collected from https://reviews.apache.org/r/25872/, it's clear that the
backoff is not aggressive enough - it is rare that we find a match for tasks within O(seconds)
if we did not immediately find a match.
> I suggest we continue to try very quickly to schedule newly-pending tasks, but apply
harsher penalties when a match cannot be found.  This would translate into 0 initial penalty
when first inserted, and perhaps a new {{-initial_schedule_delay}} default of 1 second, and
a {{-max_schedule_delay}} of 1 minute.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message