Re: 37.5% boot time reduction in Lenny is possible (recipe)
On Sun, Jun 01, 2008 at 05:40:04PM +0200, Petter Reinholdtsen wrote:
> Right. Are you talking about CONCURRENCY=startpar or something else?
> Never seen that myself, so I am curious how you get it. Could it be
> wrong init.d script dependencies in some of the packages you have
> installed? Please provide 'ls /etc/rc*.d/' for the machine in
> question, to give me a chance to reproduce this. What kind of
> hardware was this? Can you provide the bootchart graphs for the
> parallel and non-parallel boot?
Yeah I was using the CONCURRENCY= to do it.
As for hardware, well, RuggedCom RX1000 v2. That is Geode LX800, 256MB
RAM, 256MB silicon systems compact flash on the IDE port, running UDMA,
capable of about 9MB/s read.
Some parts of startup have to be done in certain orders to to
dependancies, and account for a big part of the startup time, so I am
not sure going parallel actually has much chance to help. There aren't
very many services to start. It seems the extra overhead it needed to
manage the parallel startup along with forking multiple shells to handle
them, actually added a couple of seconds to the startup much to my
surprise. Not that 30 seconds is really that bad. Turning on more
sercices might make the situation different, but I decided that there
wasn't enough potential parallelization possible to be worth the risk of
screwing up the boot in this case.
A typical PC or server probably has a better chance of getting a benefit
from parallel init.