[linux-dvb] Re: Strange buffer overruns...

[Josef Wolf <jw@raven.inka.de>]

On Wed, Dec 01, 2004 at 03:27:08PM +0100, Johannes Stezenbach wrote:

> > - I have opened 8 files (one file for each program) to write the output
> >   after I demux the contents.  Writes to the files occur nonblocking,
> >   usually in 3kb..8kb chunks per write.
> Writes to files are always non-blocking (unless you use O_SYNC or O_DIRECT)
> because they go to the buffer cache, which is written to disk later.

Are you sure?  What if you run out of buffer chache?  Without O_NONBLOCK, I
would expect to block until some buffers are freed again.

[ ... ]
> Maybe you should ask the file system gurus on lkml.

I had the impression that only gurus are allowed to post to lkml ;-)

> BTW, which chipset? For some old VIA chipsets it was well known that
> the PCI bus is blocked during IDE transfers.

lspci says:

 0000:00:07.1 IDE interface: Intel Corp. 82371AB/EB/MB PIIX4 IDE (rev 01)

so I would not assume VIA.

> You could try to increase the rate
> at which the pdflush kernel thread is woken up (/proc/sys/vm/something);
> if you use ext3 you could try to mount with commit=1.

This commit option seems to help.  With values below 3, the problem
disappears.  Thanks!

> (see linux/Documentation/filesystems/ext3.txt).

Ough, I can not find this file anywhere (not even on google).  Just
discussions _about_ the file (mostly in context with "data=journal"),
but not the file by itself.

-- 
Please visit and sign and http://www.ffii.org
-- Josef Wolf -- jw@raven.inka.de --