[conspire] quiet , and mepis 6.5

Rick Moen rick at linuxmafia.com
Wed Apr 11 19:52:49 PDT 2007


Quoting Edmund J. Biow (biow at sbcglobal.net):

> Hmmm, I Googled 'nvidia southbridge problem' & 'nvidia southbridge
> linux' and nothing grabbed me by the throat.  Maybe they just don't like
> you, Rick. :)  

Hmm, here's part of my write-up for a recent 1U rackmount server from a
possibly familiar firm we might call Dunn Microsystems, with other
identifying details trivially obscured:

  RHEL3
  RHEL3U5/64: failed on "forcedeth" ethernet driver
  RHEL3U5/64: failed on "forcedeth" ethernet driver
  RHEL3U1/64: failed on "forcedeth" ethernet driver
  RHEL3U1/64: failed on "forcedeth" ethernet driver

  RHEL4
  RHEL4U4/32: no problems
  RHEL4U4/64: no problems
  RHEL4U1/64: kernel panic because of old MPT Fusion SAS driver
  RHEL4U1/32: kernel panic because of old MPT Fusion SAS driver

  SLES9
  SLES9 SP3/64: no problems

  SLES10
  SLES10/64: no problems


  Note1: It is vitally important on RHEL4 to enable ACPI. Otherwise, a
  variety of intermittent hardware access and reliability problems will
  result, after imaging.

  Note2: It is also vitally important to know that the port that is
  assigned logical name eth0 can change between reboots. E.g., after
  connecting an ethernet cable to the port denoted 'NET0' on the back
  panel, PXEbooting, and typing "opsware_r4" at the netboot prompt, which
  results in fetching and executing via tftp an RHEL4-based OpsWare
  miniagent, when that miniagent kernel attempts to get a DHCP lease, it
  will fail until you move the network cable to the port denoted 'NET2' on
  the back panel, and retry DHCP.

  Note3: OpsWare boot label 'opsware' does not reliably work on this unit.
  You must therefore use 'opsware_r4'.


  Test unit is a 1U rackmount server, SysSN 0704BD029C, Asset tag
  #L270208. Has AMI BIOS v. 08.00.12 dtd 2006-09-08, 2 x Opteron model
  2218 @ 2.60 GHz CPUs, 3.5 GB RAM, AMD 8132 northbridge, Nvidia Nforce
  CK804 PRO southbridge, LSI Logic MPT Fusion-family SAS1064-IR SAS chip
  with MPT Fusion 6.06.06.00 BIOS dtd. 2006-03-22, no CD or DVD drive, 2 x
  Seagate 73GB SAS hard drives model ST973401LSUN72G, ATI Rage XL video, 2
  x Nvidia CK804 ethernet ports ('forcedeth' driver; usually assigned eth0
  and eth3 by the running OS) 2 x Intel model 82546EB e1000-family
  ethernet ports ('e1000' driver; usually assigned eth1 and eth2 by the
  running OS).

  It's important to note that [this model] differs from the [model it replaced],
  primarily in having the Nvidia southbridge, and two of the four ethernet
  ports being Nvidia rather than all four being Intel.


Now, Dunn's earlier version of that same model, with a _very_ similar
model designation (i.e., same but for a suffix), passed all
OS-certification tests with flying colours -- _and_ doesn't play wacky
games with which network port is eth0.  (A colleague tells me that it
also shuffles which _MAC address_ is on which physical port,
unpredictably.  Both of those things are just pathetic.  Also, the poor
state of the forcedeth driver reflects the fact that Nvidia notoriously
isn't cooperating with the Linux community.)

The earlier iteration of that model had a AMD 8131 northbridge and AMD
8111 southbridge.  Switching the southbridge chipset from the latter to
Nvidia CK804 PRO basically screwed up this unit completely.  (The
southbridge portion of a motherboard is where all the I/O ports are.)

I've seen the same thing on several other models and manufacturers:
Perfectly decent motherboard designs go all to Hell in a "Rev. B"
iteration that switches to an Nvidia southbridge.  Thus my comments.

> Say what you will about SUSE, they still support older releases better
> than a lot of folks.

SUSE's nice.  The last few releases have been particularly good on
laptops.

> Maxtor used to have a 1 year warranty, but they were taken over by
> Seagate and have changed to a 3 year warranty, not quite as good as the
> 5 year warranty you often see on Seagate's own brand.
> http://arstechnica.com/news.ars/post/20061129-8318.html

Yes, I really meant short in comparison to the main Seagate models.
I actually used to be pretty skeptical about Seagate (quite a few years
ago), and apparently so were many longtime buyers -- because they
extended their warranty out to 5 years in order to gain back people's
trust.  It pretty much worked, especially when everyone else quietly cut
their coverage periods.

> Which is why I mentioned getting a Fortron rather than whatever POS
> comes with the case below.  The least reliable Fortron apparently has a
> lower failure rate than the most reliable Antec PS, & I've even had good
> luck with Antec Smartpowers (though a friend did have an Antec explode
> on him while he was tinkering, splattering his motherboard and beard
> with goop, though it didn't do any lasting harm).

Your information may be a lot more recent than mine.  As CABAL attendees
know, I haven't bought much new hardware _personally_ in quite a long
time.  So, my prejudices in hardware sometimes are a bit antique.  

I don't really know Fortron very well, so I was probably speaking in
ignorance.  The fact that (I now see) AOpen uses them speaks well for
them, along with the fact that they tend to use 120 mm fans.  (Big fans
are much to be preferred over tiny ones, where feasible.)

> I'd also recommend getting an Antec case, but I was trying to find stuff
> that was on sale at the moment and I didn't see any good deals on Antecs
> at the moment.    Apart from noise issues, it is just much pleasanter to
> build up a well designed, solid case than some crepe paper POS. 

What you said!  

I have to confess that I really didn't appreciate that fact until I
helped my mother-in-law fix her new PC.  She'd gotten some Fry's special 
gamer-type clear-plastic lunchbox enclosure, assembled it herself, and
had all manner of problems.  She eventually took most of it back, but
kept the mini-ATX motherboard, and asked for my help.  We went to
Central Computer in Santa Clara, spotted that nice, black Antec Sonata, 
and went for it.  She's never regretting that:  It's dead-quiet,
everything _fits_ perfectly, everything's in the right place, and it just
is pleasant to work with.

I was actually inspired to replace the 120mm fans on my old VA Research
StartX MP workstation box (SunSheer mid-tower chassis) with proper
CoolerMaster replacements with decent bearings, and I'll be damned if 
that didn't suddenly get pretty quiet, too.

> The very quiet Sonata with a 450 watt PS sometimes goes for $50-65 after
> MIR, which is about what I speced for a Fortron and a no-name case. 

Damned fine deal.

> Maybe YOU interact most with the keyboard, Rick, but I never touch one. 

On the keys pretty much all the time.  The only thing that isn't set for
vi keybindings is the shell.  ;->

> Thanks, Rick, I'll try again with rsync. 

rsync (in this context) is really just a decent file copier with
sensible defaults and a reasonable command syntax, e.g., "rsync -av
source dest".  "Sensible defaults" includes doing the right thing with
weird quasi-files (device files, etc.).  

The main thing to do is boot from a live CD.  This ensures that your
source filesystems are quiescent, and that you don't have things mounted
on top of them that could give you problems.  

Having done that, you just use your favourite partition tool (me, I just
use /sbin/fdisk) to make new partitions, mkfs.[whatever] them, then
mount pairs of source & dest partitions.  (You might as well mount the
source partitions with the "ro" = read-only option; that makes it more
difficult to shoot yourself in the foot.)  Use rsync to copy stuff over.
Umount the pair.  Proceed to the next pair.  When you're done,
spot-check that things are in the right place, then install your
bootloader.  Shut down.  Remove old HDs.  Rejumper if necessary.  Then,
test-boot.  If something's still wrong, then you still haven't burned
your bridges.






More information about the conspire mailing list