[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: TODO for etch ?

On Sun, Jun 19, 2005 at 04:04:24PM +0100, Jon Dowland wrote:
> On Sun, Jun 19, 2005 at 04:16:28PM +0200, Petter Reinholdtsen wrote:
> > [Adam Majer]
> > > That could "save" a grand total of about a second. Also, during
> > > startup the bottleneck is the hard drive in many cases so starting
> > > concurrently might not speed up your boot process significantly.

> > Do you have any good references document this fact?  I've seen
> > articles documenting a speedup when things are started in parallell,
> > so I wounder where you got your information from.

> I think the redhat people who are working on this are getting the most
> benefit from a large disk read-ahead, prior to the main services being
> started. If that accounted for the majority of speed-up, maybe it alone,
> and not a fragile parallel boot process too, would be sufficient for
> most people in need of a more efficient boot.

The boot process is already fragile.  The strict ordering of boot scripts
means that any changes to the rank of one script will cause ripple effects
affecting any other scripts depending on it, which may go undetected for
long periods of time; and the available update-rc.d interface makes it more
or less impossible to automatically correct a wrongly-ordered startup script
without either losing local configuration or failing to handle systems that
don't use sysv-rc.

Switching to a sound dependency-based init system would solve the above, and
let people experiment with parallel startup or not, as they want.

Steve Langasek
postmodern programmer

Attachment: signature.asc
Description: Digital signature

Reply to: