On Mon, 11 Jan 1999, Mike Richardson wrote:
> >JJan 8 14:31:43 i3 diald[22692]: Writing error on pipe
> >/tmp/diald.info: Try again.
> >Jan 8 14:32:14 i3 last message repeated 620 times
> >Jan 8 14:33:15 i3 last message repeated 3889 times
> >Jan 8 14:34:16 i3 last message repeated 3264 times
> >Jan 8 14:34:49 i3 last message repeated 825 times
>
> Sounds to me like dialmon is not reading the pipe from diald
> fast enough. One person reported a problem where he has the link
> up almost permenantly, with long timeouts on diald; this resulted in
> lots of QUEUE entries. Dunno whether this might be the same for you.
> I'm making some changes to make dialmon better in this area, but
> it'll be a couple of weeks or so. If this is the problem a quick
> fix is to either (a) hack diald to limit the number of queue entries
> output or (b) hack dialmon to ignored queue entries after, say,
> the 20th in a batch.
On a fast-ish machine diald can easily sping through all the
connections and write the queue in a single timeslice. With
about 60 bytes per entry and the kernel having a 4096 byte
(single page) pipe buffer that means you can have just short
of 70 connections in the queue before it is guaranteed to
overflow. Web caches and servers tend to generate lots of
connections...
Hacking the monitors won't help - diald can flood the pipe
before the monitor gets a chance to drain it.
The next diald will allow TCP connections which effectively
buffer a lot more data. For pipes the only solution is to
simply not write so much data. Or hack the kernel to use a
larger pipe buffer size...
Mike
--
.----------------------------------------------------------------------.
| Mike Jagdis | Internet: [EMAIL PROTECTED] |
| 280, Silverdale Road, Earley, | Voice: +44 118 926 6996 |
| Reading RG6 7NU ENGLAND | Work: +44 118 989 0403 |
`----------------------------------------------------------------------'
-
To unsubscribe from this list: send the line "unsubscribe linux-diald" in
the body of a message to [EMAIL PROTECTED]
