[jira] [Commented] (LUCENE-5641) SimpleRateLimiter is too simple

["Robert Muir (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/LUCENE-5641?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13989623#comment-13989623 ] 

Robert Muir commented on LUCENE-5641:
-------------------------------------

Looks good, i am just a little worried about the test. For example TimeLimitingCollector's test doesnt even fail but just prints. I dont think that should block fixing this, but its just something to think about.

Maybe later we can add unit tests (like some kind of MockRateLimiter) to test the actual logic without depending on wall-clock time?

> SimpleRateLimiter is too simple
> -------------------------------
>
>                 Key: LUCENE-5641
>                 URL: https://issues.apache.org/jira/browse/LUCENE-5641
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Michael McCandless
>            Assignee: Michael McCandless
>             Fix For: 4.9, 5.0
>
>         Attachments: LUCENE-5641.patch, LUCENE-5641.patch
>
>
> I was playing with merge throttling and discovered that our
> SimpleRateLimiter is throttling far more than requested; e.g. I asked
> for 50 MB/sec merge throttling, but it throttled at more like 8
> MB/sec.
> The problem is we are calling Thread.sleep on too-small (< 1 msec)
> times; on ordinary (non-real-time) JVMs, anything less than 1 msec is
> rounded up to 1 msec.  Also, System.nanoTime() is somewhat costly.
> To fix this, I think we should aggregate the incoming byte count,
> until it crosses a threshold of enough bytes to warrant pausing.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org