Hi all,
I think I found the cause of the buffer overrun problem. My system has now been running for nearly 10 hours without a single problem, with occasional "buffer stats 0..3%" in the syslog.
It appears that the problem relies with the Ubuntu-wily HWE kernel 4.2.0-something. I downgraded to 3.19.0-something today, and since then the problem hasn't occured again. I have been load testing the whole day, with up to 5 concurrent recordings on all 4 receivers plus up to 3 live streams (xine, vnsi, and streamdev). Smooth.
Question: has anyone successfully run vdr-2.0.3 or -2.2.0 with the 4.2.0 kernel using a PCIe card? And while we're at it, a PCIe USB3 card with a harddisk attached? It seems that some kernel buffer starts choking after, like, 150 or more GB have been transferred. If that is a common problem (on any Ubuntu distro running the wily HWE kernel or on 15.10 itself) it should be mentioned somewhere - no idea where. vdr-wiki.de maybe.
I also changed the motherboard inbetween. The new one has the ethernet chip attached via PCIe (RTL8111) but my old one had not (RTL8211 / NVidia MCP97), which might be the reason why the network didn't let me down on the old MoBo. But then, I am not a kernel expert.
Phew.
Thanks to Thomas who kept kicking me. (y)
Does anyone need a K10N78 board with 4x 2 GB RAM? ;-)
(I should start using ITIL-like change management. I might have noticed it earlier. It may well be that the problems with the old USB receivers started some time after upgrading the kernel to wily HWE).
On Tue, Feb 09, 2016 at 03:15:09PM +0100, Thomas Netousek wrote:
Harald,
if
- the problem surfaces only with a new DVBs hardware card (and not
with the old one) and
- it only appears when you have traffic both on the DVB card and on
the network interface and
- the problem only goes away when you cold boot
then this makes me think that you have a problem with the PCIe bus.
I trust you have updated the mainboard BIOS to the latest level, right ?
Can you please elaborate on the hardware
- mainboard make and model
- schematic of the mainboard and how network card, disk controller
and DVB card are attached
Please run an iperf test (bi-directional) to measure the network performance when newly booted and when "the problem" has surfaced.
And I would guess that you can trigger the problem more easily if you run iperf, create disk activity (run a random access read-only disk test ?) and run the remote video client at the same time.
Btw, there are LEDs on the Digital Devices card, do they change between the "good" and the "bad" state ?
Thomas
vdr mailing list vdr@linuxtv.org http://www.linuxtv.org/cgi-bin/mailman/listinfo/vdr