High cpu utilization with raw-video(UYVY) network streaming pipeline

[Tim Allen]

Further to the performance improvements I saw for udpsink etc, using the
later glib and gstreamer source, I'm now looking at the equivalent stream
for receiving. 

Unfortunately I see a worse performance for 1.x again. I simplified my
pipeline to just "udpsrc - queue - fakesink" for comparison between
gstreamer 0.10 and 1.x. I can see from watching /proc/net/udp that for 0.10,
the (32MB) UDP rx queue regularly returns to 0 and never fills, but for 1.5,
the queue fills in a few seconds, and packets are dropped. I also see one of
the two enabled cores hitting 100% loading.

Top shows:

GStreamer 0.10.36
  PID USER      PR  NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND
 8841 ubuntu    20   0   34424   3304   2112 R 83.3  0.2   0:25.12
queue0:src
 8842 ubuntu    20   0   34424   3304   2112 R 52.1  0.2   0:14.98
udpsrc0:src
 
GStreamer 1.5.0 (GIT)
  PID USER      PR  NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND
 8831 ubuntu    20   0   43632   3892   2616 R 97.7  0.2   0:18.68
udpsrc0:src
 8830 ubuntu    20   0   43632   3892   2616 R 59.1  0.2   0:09.15
queue0:src

I notice that queue/udpsrc loading balance changes between the versions. I'm
continuing to debug and hopefullly profile this some more, together with
looking at the latest source, but wondered if there are any more
suggestions, or whether the improvements already made should have helped, or
are still needed here.

Any guidance greatly appreciated.
Thanks,
Tim.



--
View this message in context: http://gstreamer-devel.966125.n4.nabble.com/High-cpu-utilization-with-raw-video-UYVY-network-streaming-pipeline-tp4669784p4670888.html
Sent from the GStreamer-devel mailing list archive at Nabble.com.