Mailinglist Archive: opensuse-factory (710 mails)

< Previous Next >
Re: [opensuse-project] Re: [opensuse-factory] The road to systemd for openSUSE 12.1
  • From: Kay Sievers <kay.sievers@xxxxxxx>
  • Date: Wed, 15 Jun 2011 11:48:56 +0200
  • Message-id: <1308131337.1488.12.camel@mop>
On Wed, 2011-06-15 at 09:53 +0200, Frederic Crozat wrote:
Le mercredi 15 juin 2011 à 09:08 +0200, Carl-Daniel Hailfinger a écrit :
Am 11.06.2011 21:53 schrieb Greg KH:
[...] Because what we have right now sucks.

Seriously, it does, it was great for the 70's and 80's when things were
static, but now, it makes absolutly no sense whatsoever. Linux has been
evolving to support this type of dynamic, use only what you need when
you need it, type of a system for a very long time now, and this is just
one piece of that progression that has been needing to change for a very
long time.


So if I'm installing a web server machine with apache, it will boot
quickly, but the first access to the server will take almost forever
because nobody connected to port 80 before and thus apache wasn't
started? Or does "use only what you need when you need it" not apply
here (and if not, why)?

Socket activation is mainly used to simplify service startup and to
avoid to express any dependencies, and not about on-demand start.

On a common systemd setup, there are almost zero dependencies to
resolve, because all communication channels between services are
established before any service is ever started. The kernel will just
suspend the requesters until the real service is running.

The same logic applies to filesystems access when systemd's automounter
is used.

Most common services are started just unconditionally, like they have
been started since forever, regardless if there are traffic on the
socket or not.

Looking at http://0pointer.de/blog/projects/socket-activation.html it
seems that systemd socket activation adds an additional overhead of a
few context switches for every single connect. How big is the
performance degradation for short-lived TCP and UDP connections in that
case?

Are you running apache behind xinetd ATM ? I don't think so. So I don't
see why you would use socket-activation for apache server.

Right, if you run a real webserver, you don't run it inetd mode. Also,
the overhead for native/non-inetd systemd socket activation is zero.
Apache listens itself, it only gets the first connection passed, if
configured to run with socket activation.

SShd is a fine example, and there are not many, to run by default from
systemd's inetd mode. On most workstations nobody ever logs-in over ssh,
and it can be started on demand just fine. If you run a server, where
users log-in over ssh all the time, you wouldn't run sshd in inetd mode
there, but just enable the service unconditionally, so it is immediately
ready when users connect.

Kay

--
To unsubscribe, e-mail: opensuse-factory+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: opensuse-factory+help@xxxxxxxxxxxx

< Previous Next >
This Thread
Follow Ups