On Jan 11, 2015, at 11:05 AM, Valeri Galtsev <galtsev at kicp.uchicago.edu> wrote: > On Sun, January 11, 2015 11:22 am, Sven Kieske wrote: >> >> On 11.01.2015 03:42, James B. Byrne wrote: >>> What does systemd buy the enterprise that sysinit did not provide? >>> >> systemd has it's ugly downsides, but it >> _does_ provide much needed features. > > I don't care that _laptop_ with systemd starts 3 > times faster - it's brilliant when you have to start it right on the > podium few seconds before giving your presentation. What about all those poor enterprise people who have been arm-twisted into agreeing to SLAs? If you’ve agreed to provide five nines of availability, a single reboot in the old BIOS + hardware RAID + SysV init world could eat most of the ~5 minutes of downtime per year you’re allowed under that agreement. EFI + software-defined disk arrays + systemd might cut that to a minute, allowing several reboots per year. Until we start to see hot-upgradable Linux kernels in mainstream distributions, I’d say that does amount to an “enterprise” feature. You can extend this argument to four-nines, where you only get 4 minutes of downtime per month. Looking through the centos-announce list archive, there seems to be roughly one kernel-* RPM change per month. Do you really want to burn your entire downtime allowance on that?