#359: CPU spike on metadata updates
------------------------+---------------------------------------------------
 Reporter:  humanclay   |       Owner:  admin       
     Type:  Bugs        |      Status:  new         
 Priority:  10          |   Milestone:  NEAR FUTURE 
Component:  Liquidsoap  |     Version:  1.0.beta+svn
 Keywords:              |         Mac:  0           
    Linux:  0           |      Netbsd:  0           
    Other:  1           |     Freebsd:  1           
------------------------+---------------------------------------------------

Comment(by mrpingouin):

 Okay, everybody calm down, it was just me screwing up: I was just
 observing a bug I recently introduced in the time measures (liquidsoap was
 running fine but displaying misleading figures).

 Now, I've fixed it in r7680 and could run tests that make sense. As
 expected, only vorbis encoders are affected by metadata (for vorbis,
 metadata means new track; for other formats it is simply dropped). Still,
 it is hard to create a significant latency with a vorbis encoder on my
 laptop: the maximum stress is to send a metadata insertion every 0.1
 seconds, and that leads to no more than 3 or 4 cycles latency.

 I have attached a new test and the results on my server (Celeron 1.2GHz
 running at 50% of its speed). The test sends 10 metadata insertions at
 0.1s intervals, then waits 3s, and loops. I think it's way more stressful
 than any reasonable use case. With %wav, I get no latency at all
 (server_wav_10_every_3.png). With %vorbis, I get about 25 cycles = 1s
 latency at every round of metadata updates (server_vorbis_10_every_3.png).
 As expected if I only send strikes of 5 metadata updates the latency gets
 better (server_vorbis_5_every_3). Notice that there is no overall drift of
 the clock.

 In the end, I'm not sure that this is such an unreasonable figure. On your
 machine it might be worse: please send us some nice pics! Your processor
 may be less powerful, and also might not have a FPU. Also, as said above,
 aotuv might have a different impact than libvorbis; it would be really
 nice if you could send a comparison.

-- 
Ticket URL: <http://savonet.rastageeks.org/ticket/359#comment:4>
Savonet <http://savonet.rastageeks.org/>
Let's program our stream !
------------------------------------------------------------------------------
Start uncovering the many advantages of virtual appliances
and start using them to simplify application deployment and
accelerate your shift to cloud computing.
http://p.sf.net/sfu/novell-sfdev2dev
_______________________________________________
Savonet-trac mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/savonet-trac

Reply via email to