I have been playing around in this arena a little and collecting data on
my equipment.
I started off in the Ultra66 world with a Promise Ultra66 controller an
IBM-DTLA-307010, and an IBM-DTLA-307045 on an old P133 w/ Mandrake 7.0.
After drive optimizations I was able to get ~12 MB/s off of the 45 GB
drive when I ran hdparm -t on it.
A few months later I got my hands on a BE6-II m/b w/ an htp366 controller
onboard. I could never get my 45 GB drive to work right with this m/b
under Linux and eventually gave up on the board. I recently noticed that
the 2.4.1 pre11 had a fix that may cover this problem, but I don't have
the board now.
What I am currently using is an ASUS K7V m/b w/ Promise Ultra100 and
Ultra66 controllers (off board). I noticed that with the 2.2.17 kernel
that comes with mdk 7.2 with an IBM-DTLA-307045 on ide2 and ide3 that if
autotuning is on then the machine will hang on boot. I found that it
didn't matter which controller I used (Promise Ultra66 or Ultra100).
Also, if I removed the hard drive from ide3 and had drives on ide3, ide4,
and ide5 then the system would come up. I got around the problem by
optimizing after boot. I also noticed that if I unmask the IRQs under
heavy disk activity on all four disks hanging off the system then the
machine would hang. I didn't didn't try to narrow it down from that
though it may be a problem with just one or two disks. It may even be the
same problem that gets me when I have autotuning turned on. Also I
noticed that on the second Promise controller (this case the Ultra66) that
when I ran hdparm -t after optimizing under 2.2.17 that I would only get
~24 MB/s. When the Ultra66 controller was by itself it was able to get
35.56 MB/s. The cure to these problems is in the 2.4.0 kernel and up so I
didn't really see a point. The main discrepancy that I saw in 2.4.0 was
that I would get 35.75 MB/s off of the drives on the Ulta100 controller
while I would only get 35.56 MB/s off of the Ultra66 controller. This was
fixed with a BIOS update to the Ultra66 controller. With a three disk
software RAID level 0 under 2.4.0 w/ 32KB stripes I was able to achieve
~50 MB/s when I ran hdparm -t.
If you ever go after software RAID here is what I noticed so far. I heard
reports that various RAID levels on 2.2.17 work on SCSI in software, but
when I testing things out on my system w/ IDE only it was just screwed up
all over. I couldn't get RAID 1 to work (got it working on 2.4), then I
had massive data corruption problems with RAID 5. Plus there where those
controller issues. On 2.4.0 everything seemed to work except that I would
get some corruption while running a software RAID 5 array in degraded
mode. On 2.4.1 the problems that I was having with RAID level 5 in
degraded mode went away. (I noticed that the 2.4.1 release included some
RAID level 5 fixes and cleanups according to the release notes.) Even
though I have tried just about everything on 2.4.1 that I tried on earlier
kernels, there are still a few things left for me to do, so I can't give a
full report on this matter yet.
The summery of this goes as follows. I find the IBM-DTLA-307045 to be
blazingly quick, especially when combined with Reiserfs on a single
drive. For an IDE drive the one seems to perform quite well in all
aspects. If you are looking to have more than one drive hanging off of
your controller and you are not running 2.4.0 or above then you may see
trouble with the Promise controllers. Until mdk 8.0 comes out your best
bet from what I can tell is to pick up a Promise Ultra100 and only have
one fast disk hanging off of it. (Maybe try a different controller if you
want more now without hassling with kernels.)
Cecil Watson wrote:
> Hello Everybody,
>
> Hi Dr. Nick! I think of getting a Promise UDNA 100 controller (time to
> return my 66 to my BeDVBox). I few weeks ago there was a discussion
> about 66/100 drives not living up to specs. So I'm looking for
> reccomendation on a drive. Brand and model if you please. Thanks in
> advance,
>
> Cecil
