But now for something completely different, how to shorten these long delays in kernel 2.6?
- From: Jan Panteltje <pNaonStpealmtje@xxxxxxxxx>
- Date: Tue, 28 Feb 2006 14:56:22 GMT
I have a number of situations where there is video input piped via some programs and then to a display.
Delays for a live event to become visible:
USBwebcam -> driver_module -> v4l-userapp(display1) -> pipe -> ffplay(display2)
0 1second 6 seconds
This stays the same long delay when setting buffer sizes to 0 and using fflush() in the applications.
I ended up in /proc/sys/vm/
the 2.6 kernels no longer have bdflush, but 2 threads are running it seems:
cat /proc/sys/vm/nr_pdflush_threads
2
I am not very confident of tweeking params here in /proc/sys/vm/
So the question is, before I ... this system, WHAT can I do to make the 6 seconds go away?
The 6 seconds delay is also there if I use an other video source, for example ethernet port (so not USB).
I suspect Linux caches the data somehow somewhere in the pipe, strange is that the delay
is always the same, even at low (2fps) frame rates and small display sizes.
So what shall I tweek, and wil lit even make a difference?
l /proc/sys/vm/
-rw-r--r-- 1 root root 0 2006-02-28 15:53 vfs_cache_pressure
-rw-r--r-- 1 root root 0 2006-02-28 15:53 swap_token_timeout
-rw-r--r-- 1 root root 0 2006-02-28 15:53 swappiness
-rw-r--r-- 1 root root 0 2006-02-28 15:53 page-cluster
-rw-r--r-- 1 root root 0 2006-02-28 15:53 overcommit_ratio
-rw-r--r-- 1 root root 0 2006-02-28 15:53 overcommit_memory
-r--r--r-- 1 root root 0 2006-02-28 15:53 nr_pdflush_threads
-rw-r--r-- 1 root root 0 2006-02-28 15:53 nr_hugepages
-rw-r--r-- 1 root root 0 2006-02-28 15:53 min_free_kbytes
-rw-r--r-- 1 root root 0 2006-02-28 15:53 max_map_count
-rw-r--r-- 1 root root 0 2006-02-28 15:53 lowmem_reserve_ratio
-rw-r--r-- 1 root root 0 2006-02-28 15:53 legacy_va_layout
-rw-r--r-- 1 root root 0 2006-02-28 15:53 laptop_mode
-rw-r--r-- 1 root root 0 2006-02-28 15:53 hugetlb_shm_group
-rw-r--r-- 1 root root 0 2006-02-28 15:53 dirty_writeback_centisecs
-rw-r--r-- 1 root root 0 2006-02-28 15:53 dirty_ratio
-rw-r--r-- 1 root root 0 2006-02-28 15:53 dirty_expire_centisecs
-rw-r--r-- 1 root root 0 2006-02-28 15:53 dirty_background_ratio
-rw-r--r-- 1 root root 0 2006-02-28 15:53 block_dump
.
- Follow-Ups:
- Prev by Date: Re: libc/printf bug
- Next by Date: Re: Trolltech QT license question
- Previous by thread: libXft2
- Next by thread: Re: But now for something completely different, how to shorten these long delays in kernel 2.6?
- Index(es):
Relevant Pages
|
|