“Last year leading up to the release of Ubuntu 7.04 and Ubuntu 7.10 we had published several articles looking at various aspects of this desktop Linux distribution. These articles had varied from looking at Ubuntu’s power consumption for the past six major releases to presenting the visual history of Ubuntu and how its graphics have evolved since Ubuntu 4.10. With Ubuntu 8.04 shipping in just two months, we are once again looking at Ubuntu from several points of view. In this article, we are looking at Ubuntu’s boot performance for the past five releases through the use of Bootchart for measuring its boot time, disk throughput, and the running processes.”
I was hoping for a little more substance when I read the article, instead it was basically a chart with little real insight behind the numbers.
Ubuntu boots faster with less processes starting at boot. News at 11.
I’d really have been interested to see some granular analysis of upstart. I can see the value of an event-based service mechanism from the POV of services only being started, or activated on-the-fly, as required, at least as one example, but from a booting perspective, I’ve been curious to know if there really is a tangible benefit versus a carefully crafted sysvinit config with parallel service starting.
I think that disk throughput is the bottleneck that everyone is trying to work around, but I don’t know how much more trimming there is to do. Certainly it makes sense that something like upstart may not start CUPS if there are no printers available or network service started, but then a standard startup could also push CUPS to start after X etc. has been loaded, giving the Microsoft-esque perception of quicker boot where services continue loading even after the login display has appeared.
Aside from taking the time to tweak my own default services, I’m fairly lazy and so was hoping somebody else had made the effort and done the proper legwork to see if it really makes a difference…
Well s far as know Upstart really isn’t being used in it true form. Right now they are still using the sysv compat layer and init scripts. So at this point what you suggested might be exactly what is happening. Hardy was supposed to be the version where they started migrating packages to use upstart in earnest but I guess that got postponed.