[ale] LILO laying low

[Benjamin Scherrey]

I just noticed that neither my reply nor your response to it made it on the group
(perhaps the list admin might rethink the policy of not populating the reply-to field
on ale mail???) so I've left everything intact on this email in hopes someone will be
able to answer your new question better than I.

Courtney,

    I've never run into the question of partitioning on cylinder boundaries before but
I've long since used disk druid rather than fdisk for my linux installs so that's
probably why. My guess is that if fdisk allows it then its probably ok but simply a
warning that some systems may not be able to work under such an environment. I can see
why some system software would be easier to write if it counted on a per-cylinder level
of granularity for its partition tables.

    That said, the initial way to determine your cylinder size is to multiply the
sector size in bytes by the number of sectors per cylinder. This information should be
available about your drive. As long as your partition sizes are equally divisible by
that value you should have partitions allocated on a cylinder boundary. That said,
however, there is also the concept of logical vs. physical drive configurations so that
pretty much means all bets are off unless you get an answer from someone who's more
hardware inclined than myself.

    On allocation of swap and proc, again, I expect that the sysadmin types who roam
this list will have better answers (thus reasons) than I but I'll tell you what I do,
FWIW. I always first setup the partition for /boot which is typically 8MB (cause I have
lots of kernels installed and my lilo list gets kinda big). Then I allocate my swap
partition. The rule of thumb used to be allocate double your RAM space but that was
when the linux kernels weren't as efficient with file buffers and memory management and
many systems had less than 32MB RAM. Now I'd say to allocate at least 32MB or equal to
your RAM space but not more than 128MB. Frankly I compile kernels and the entire gcc
environment (along with some of my own heavily templated, thus RAM consuming, code) all
the time on my box with 256MB RAM (128 swap) and have never seen it go over 10MB swap
space usage. Its usually sitting at about 3MB swap usage. Beyond that I typically
allocate the rest of the drive to / so proc, var and usr are all on the same partition.
This is fine for a user workstation but a webserver or other server class box should
probably split this up for performance, security, and potential corruption issues.

    regards & good luck,

        Ben Scherrey

RHS Linux User wrote:

> Benjamin Scherrey wrote:
>
> > This most commonly happened to me when putting linux on a big drive and not
> > having a seperate, small (say 8MB or so) partition that was the first one on the
> > drive be mounted as /boot. This way lilo is guaranteed to only have to access the
> > first part of the drive. Otherwise, you might get bad luck and have your boot
> > kernel file sectors scattered onto a later part of the drive which lilo cannot
> > access. This is why a seperate boot from floppy allows you to access the drive,
> > the kernel's already loaded. So.. how is your system partitioned.
> >
> >     good luck & regards,
> >
> >         Ben Scherrey
> >
> > Courtney Thomas wrote:
> >
> > > Greetings !
> > >
> > > When I try to bootup, at the prompt I get LIL-.....
> > > and no bootup.
>
> Ben,
>
> Thanks for your reply.
>
> All is working again, but I do have a related question.
>
> When I fdisk the drive I see messages to the effect that.......partitions are not
> ended on cylinder [I think that's what it said] boundaries. Is this imperative ?
>
> 1-If not imperative, so what ?
> 2-If so, how should I layout the drive [8 Gig] to insure that each partition ends on
> a cyl boundary ?
> 3-Also, how should swap and proc implemented?
>
> Since I am going to start over, since receiving your /boot info, I might as well
> straighten-out everything else.
>
> I hope all this is not too importunate. If so, disregard. I understand.
>
> Appreciatively,
>
> Courtney

--
To unsubscribe: mail majordomo at ale.org with "unsubscribe ale" in message body.