Re: Sparc Debian?

[Johnie Ingram <johnie@netgod.net>]

"Michael" == Michael Solomani Mifsud <solomani@aic.net.au> writes:

Michael>  Does Debian make a version of Linux (UNiX?) for sparc
Michael> machines?  If so, what kind of preformance comparison between
Michael> solaris 2.5 and SunOS?  Thanks.

Funny you should ask -- I just installed it the other day.

The Sparc version is still under development, but I did manage to get
it to work.  In fact I put it on a mission-critical production system,
with some 1500 users, and most never noticed a change.

I had great faith in Linux and Debian, obviously, from long
experience.  :-) 

Before this it ran the venerable SunOS 4.1.3; I can't compare it to
the newer Solaris.  But after running it for two days in a
less-than-ideal configuration, it seems comparable.

There were some problems at first with it running out of memory until
I added a 64MB swapfile to the 64MB RAM.  The swapfile and the NFS
root probably slow it down.  An honest comparison can't be made until
I install it just like SunOS was, permanent-style -- separate /,
/usr/, /var, swap, and two /home partitions, spread across 3 SCSI
disks.

But right now the system load bounces around 1.3 with 28 users using
pine, and sendmail MTA, if that's any basis for comparison.

You can hit "delete" now -- what follows is an account of my experience.




-- clip here --

Date: Mon, 25 Aug 1997 06:07:14 -0400


If anyone's interested in my endeavor to secretly replace SunOS 4.1.3
with Linux on a SparcStation 20.

First I FTPd the kernel and debian-sparc-root and installed it
according to the doc file (which curiously, referred to it as
sparc-debian-root).  Then I configured bootpd and did a "boot net" --
which failed of course.

I had to follow the HTML links and study the manuals before realizing
that bootpd doesn't do arp.  Then there was a few minutes of tension
where rarp didn't seem to be available at all, but it turns out that
kerneld just wasn't autoloading it.

Then there ensued the usual problem with tftpd being off in
hosts.allow.

Now the kernel booted, repeatedly.  The ftpd daemon reported a
successful fetch every 10 seconds.  It took a half hour to determine a
problem with an IP address was causing this -- the going theory on IRC
was that the kernel itself was compatiable with sun4m.  (I installed
the RedHat kernel briefly to make sure this wasn't the problem.)

Now it worked, and I was never so happy to see the "color or mono"
dialog box.  I entered dinstall, went straight to the shell option (I
didn't yet realize VCs worked on sparc), and tried to fdisk.

There was no fdisk anywhere, or dmesg -- the two things i needed most.

Undaunted, I went to the NFS server and ar x'd the fdisk.deb and
tarred it into place.  Then since I was at it, I did some other
packages too.  Returning to the sparc I found this broke a bunch of
things, like "ls", for some reason.

The #debian channel on IRC suggested i just unpack the base.tgz and
mount THAT, which worked better.

I tinkered around for a while, switching between SunOS 4.1.3 and linux
to make absolutely sure the disk I was about to wipe was "/dev/sd2".
(I have three disks, two of which are identical 2GB -- one with vital
/home data, the other empty.)  Finally after checking the SCSI IDs I
guessed that SunOS boots from /dev/sdc and that "/dev/sd2" was
/dev/sdb.

When it came down to the actual partitioning it turned out I had no
idea how to repartition a Sun disk.  When fdisk started saying it was
recommended that /dev/sdc3 be a "whole disk" partition, I basically
gave up and decided to just mke2fs what was already there -- a 2GB
partition linux called /dev/sdb7.

At this point it was smooth sailing until I decided to try runlevel 2.
Turns out there's a /etc/init.d/network in base.tgz that kills the
RARP configuration the kernel did, so the system hangs a split second
after going multiuser.

I also found I had no /etc/inittab, which init.d didn't like.  And I
had to edit /etc/fstab to put a NFS entry there for /.

Now I had multiuser linux going, and an all-but empty partition, a
familar scenario.  Crossing my fingers, I was happy to find that
mount -t ufs /dev/sda7 actually worked -- not only did someone think
to compile ufs in, this also meant I hadn't trashed the good disk.

I started a copy of the 1.8-odd GB of /home data to the linux
partition, which even now -- six hours later -- hasn't finished.
Yet linux insists they are all "10.000MHz FAST SCSI-II" drives.
Hmmmmm.

In the meantime I set about installing Debian/Sparc.  Turned out
crucial packages weren't there, like the libc6 that grep and sed
seemed to want.  Nor were some packages I need to have before morning:
apache, bind, sendmail, proftpd, etc.

So I set about installing the development environment.  Turns out that
even with automake and gettext installed, I couldn't build dpkg
1.4.0.19.  The provided dpkg 1.4.0.8 can't unpack source, so I've had
to put up with "zcat ../package*diff.gz | patch -p1" all night.

This is where I'm at now, running "dpkg-buildpackage -B" on stuff I
absolutely need, like bitchx-bin and linux_logo.  :-) Most of it is
going smoothly.  Emacs is the notable exception -- it seems to have a
Source-depends on itself.  And with my bogomip rating (50), load level
(4), and disk throughput (pathetic), every attempt to "debian/rules
binary" takes over 20 min.  :-(

Incidentially linux_logo source seems to need a sparc patch.

Once everyting is working I'll go back and do everything for real,
with libc6 instead of libc5.


---------------------  PGP  E4 70 6E 59 80 6A F5 78  63 32 BC FB 7A 08 53 4C
 
   __ _    Debian GNU         Johnie Ingram <johnie@netgod.net>      mm   mm
  / /(_)_ __  _   ___  __        "netgod"     irc.debian.org          mm mm
 / / | | '_ \| | | \ \/ /                                             m m m
/ /__| | | | | |_| |>  <      Those who do not understand UNIX       mm   mm
\____/_|_| |_|\__,_/_/\_\     are doomed to repeat it, poorly.       GO BLUE


--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-user-request@lists.debian.org . 
Trouble?  e-mail to templin@bucknell.edu .