[ http://dev.sourcefabric.org/browse/LS-376?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Romain Beauxis updated LS-376:
Priority: Major (was: Blocker)
Ok, I have tested locally here (slow netbook cpu).
I notice a peek when passing new metadata, which prob. corresponds to closing of old vorbis encoder and opening of a new one.
However, the peek does not seem to be as bad as described in this ticket.
Since this ticket is rather old, I downgrade it to major and wonder if it can be reproduced.
Finally, in case of systems with slow cpu, it should be possible when outputing to icecast to use the icy_metadata mechanism to change metadata without restarting the encoder, which should avoid most of those peeks.
> CPU spike on metadata updates
> Key: LS-376
> URL: http://dev.sourcefabric.org/browse/LS-376
> Project: Liquidsoap
> Issue Type: Bug
> Components: Liquidsoap
> Affects Versions: 1.0 beta
> Reporter: anonymous
> Assignee: Romain Beauxis
> Priority: Major
> Fix For: NEAR FUTURE
> Attachments: spikes.liq, vorbis_tricky_reset.patch
> The primary machine that I have experienced this issue on is:
> VIA C7 Processor 1000MHz with 512MB of RAM
> FreeBSD 7.1 i386.
> Liquidsoap 0.9.2-2 using OSS input module
> aotuv 20090303 libvorbis (http://www.geocities.jp/aoyoume/aotuv/)
> I have run these tests using a single vorbis stream. I have found that regardless of the quality level (-2.0, -1.0, 0), the skips can occur during metadata changes. In some situations I have found that the stream continues to loop the same audio over and over again and does not get unstuck until liquidsoap is restarted, which is really bad.
> I was able to make the skips occur by rapidly copy and pasting "meta.insert title='metadata'" over and over again into the telnet session. When skips occurred they were quite noticeable.
> I noticed this in top:
> Under normal conditions:
> CPU: 35.2% user, 0.0% nice, 2.2% system, 3.7% interrupt, 58.8% idle
> During metadata updates:
> CPU: 98.9% user, 0.0% nice, 0.7% system, 0.4% interrupt, 0.0% idle
> (It took me awhile to be able to copy-paste this from top as the CPU quickly spiked up and then went right back down)
> Interestingly enough, I performed the same tests on an Intel(R) Pentium(R) 4 CPU 3.00GHz with 1GB of RAM, again running FreeBSD 7.1 i386. The skips also occurred on this machine but they were far less noticeable. The audio just had a little "blip" in it on metadata updates.
> So it appears that there is a quick spike in CPU usage that is happening on metadata updates, which is exemplified on slower hardware.
This message is automatically generated by JIRA.
If you think it was sent incorrectly contact one of the administrators: http://dev.sourcefabric.org/secure/Administrators.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira