Or, to put it in other words, the only major distros that currently don't ship systemd as the default init are Debian and its derivates
So you don't think that Red Hat RHEL, Suse Enterprise, Centos, and Oracle Linux are major distros then? So far as I know, none of them are shipping Systemd. Look at the distros that people are using to run their businesses. They're all using either System V or Upstart. Even Fedora doesn't use Fedora on their own servers.
Systemd won't really have been put to the test until one or more of the above starts using it as the default init they ship to paying customers for use on servers. Debian and Ubuntu differ from Red Hat and Suse in that they don't have separate "paid" and "free" versions. If Fedora crashes spectacularly or won't boot, Red Hat just shrugs and carries on. If Debian or Ubuntu have problems, then their actual customers are directly affected.
Systemd is supposed to make life easier for distro packagers. From the end user's point of view however, there's no tangible benefit. I could quite easily understand if Debian decided they didn't want to be the guinea pig on this one.
I wonder what approach Ubuntu will take? Since upstart was developed by Canonical, they may not abandon it for political / personal reasons.
I think they said they don't have any fundamental objections to it, but they had other more important things on their plate at the moment. Right now, their focus is on Unity for the desktop and sorting out their cloud stack for servers. Their next release will be an LTS with five years of support, so they are concentrating on polishing what they've already got rather than ripping out and replacing all the plumbing.
I got the impression however that they're waiting for someone like Red Hat or Suse to go first with Systemd and then seeing how it works out for them before making any decisions for themselves. What Debian decides will play a big part as well, as most of Ubuntu's packages are just pulled straight from Debian. Upstart was designed for backwards compatibility with System V while Systemd isn't,
Phoronix Comparison wrote:
While some Phoronix readers and Phoronix Forums users thought that Fedora's increased usage of systemd would give it a lead in boot performance plus all of the other upstream optimizations and improvements made by the Fedora / Red Hat engineers, this was not the case in comparing Fedora 16 and Ubuntu 11.10.
That's what I was expecting too give that Pottering, in his LO interview, argued about start-up processes having to wait for other process to complete in the traditional boot and with upstart.
The start-up performance of upstart seems slightly better in these benchmarks: Ubuntu 11.10 vs. Fedora 16: Boot Speed, Power Consumption
No enough that anyone should care and not enough, by itself, to justify ubuntu switching to something slower
The assumption of faster boot times was based on the idea that you will somehow gain by not starting up daemons that you don't need, or by not waiting for daemons to start. However, the actual measurements were:
Bootchart in Fedora measured a start time of 32.72 seconds. Ubuntu 11.10 had a reported boot time of 32.40 seconds on the same hardware.
When running the Lenovo ThinkPad T61 with its Intel Core 2 Duo T9300 CPU, 4GB of RAM, and 100GB Hitachi 7200RPM SATA HDD, Fedora 16 booted in 28.23 seconds. Ubuntu 11.10 meanwhile booted in 22.88 seconds for the Intel Core 2 Duo notebook.
There were some developers (from Intel, I think) working on boot times for netbook/tablet applications, and according to them the real problems are:
Probing the hardware to see what's installed so you know what drivers you need, and
For the first one, the solution is to hard code the driver list, which is really only practical for things like tablets where the user can't practically upgrade the hardware.
For the second, what you really need to do is to very carefully control the order in which you load things, so that you can overlap disk I/O operations with CPU intensive operations. That's the exact opposite of what Systemd does (it does things randomly).
Poettering seems to have backed away from the improved boot time claims and focuses now on the simplified init configurations. It is possible that Systemd might eventually make life easier for distro packagers, but I would be surprised if end users see any direct tangible benefits.