On Fri, Oct 01, 1999, Markus Stumpf <[EMAIL PROTECTED]> wrote:
> On Fri, Oct 01, 1999 at 05:26:38PM +0100, Fred Backman wrote:
> > * inetd is unreliable under high loads. It cuts off service for 10 minutes
> > if it receives ``too many'' connections in 1 minute.
> > * inetd does not provide effective resource management. It will happily use
> > up all your memory if you are running a popular service.
> > * inetd has trouble with sudden bursts of activity. Its listen() backlog is
> > typically only 5 or 10 and cannot be raised.
> >
> > Where can I find further documentation and proof or witnesses of this
> > actually being true?
>
> How about
> $ man inetd
>
> On my FreeBSD system I find:
> -R rate
> Specify the maximum number of times a service can
> be invoked in one minute; the default is 256.
> However this can only be done "globally" for all services and you cannot
> say "use 255 for smtp and 5 for all other services".
>
> If you use ulimit (or whatever it's called in various shells) before you
> start inetd you can set certain limits, but they apply for all services.
> You cannot specify selective limits for individual services.
>
> For the listen() backlog you probably have to consult the source code of
> inetd for your system.
>
> With tcpserver you can fine tune each parameter for each service
> individually. For most services started from inetd this may not be needed,
> for some it may be essential (smtp is a good candidate) and then tcpserver
> is your friend.
Atleast with Linux's inetd:
-q queuelength
Sets the size of the socket listen queue to the specified value.
Default is 128.
[...]
The optional ``max'' suffix (separated from
``wait'' or ``nowait'' by a dot) specifies the maximum number of server
instances that may be spawned from inetd within an interval of 60 sec-
onds. When omitted, ``max'' defaults to 40.
Both work for me fine. I dunno what people's big gripes against inetd
are. It works damn well for almost every service that doesn't have high
loads. In 99% of people's cases who use mail, then won't get more than
40 hits a second, and if they do, increase it. I run it with a 10000 hit
maximum on a relatively busy mail server (close to 100,000 incoming messages
a day)
The only thing it can't do is set environment variables based on
source IP address (like tcpserver), but I don't allow relaying through
that machine anyway so I don't need it.
JE
