Martin Apel wrote:
>
> Hi,
>
> I investigated the problem regarding estimate problems on a host with
> many filesystems a bit. I think I have found out what goes wrong.
> It's not the size of the datagram, it's the maximum size of a pipe's buffer.
> The sendsize processes all seem to be stuck waiting for a lock when
> trying to write their results to stdout.
> Until that time they have written a little more than 4000 bytes to the
> pipe to the master sendsize process.
I've grepped the linux kernel sources a little and I found:
#define PIPE_BUF 4096 /* # bytes in atomic write to a pipe */
in /usr/src/linux/limits.h. This coincides with your observation of
"a little more than 4000 bytes". How about increasing PIPE_BUF and
recompiling the kernel?
--
Regards
Chris Karakas
Don´t waste your cpu time - crack rc5: http://www.distributed.net
