Andre Hedrick wrote:
> > I have a system with 12 udma disks, and in practice it's been a
> > nightmare. Once you start using more than two interfaces the
> > auto-tuning code in the kernel ceases to work correctly, so you
> > have to "tune up" each disk after
>
> The tuning code fails?????? News to me.
> I have several reports that three (3) Ultra33's or Ultra66's work fine,
> and get correctly tuned at init.
This was based on advice you gave me, but at that point I was using the
2.2.3 version of your unified ide patch. I'm now using 2.2.7... Raid patch
sets have limited me from upgrading further (until today, when a new raid
patch was released).
We spoke back and forth, and then you suggested:
> Otherwise, compile with CONFIG_IDEDMA_AUTO=n
> Calling ide2=dma ide3=dma ide4=dma ide5=dma
> After boot call ::
> hdparm -X34 -d1 hda
> hdparm -X34 -d1 hdb
> hdparm -X34 -d1 hdc
> hdparm -X34 -d1 hdd
Which has worked fine for me. I haven't done any experimentation since
then, as once it was working I didn't really want to screw with it. But
since we have a 2.2.10 raid release now, I will grab the latest unified ide
patch and see how it goes without CONFIG_IDEDMA_AUTO=n and the kernel
parameters.
> I would like to see a failure report on mistuning, since I have verified
> personally that I can make three promis ultra33's run in ata-33 mode
> correctly without post boot correction.
I don't mean to attack your code in anyway. It's just that this was my
impression, as you had suggested this as a fix for me.
After re-reading my post, I definately conveyed the wrong impression. It
was late, I'm sorry. When I said that "In practice, my experience with
large ide-raid has been a nightmare" I did not really mean the linux ide
code, which has been super stable since it's first install. The reason I
regard my large ide raid system to be "a nightmare" is due to cable lengths,
CRC errors on the ide bus due to cable lengths and even just cable routing
issues (don't stretch an ide cable and expect it to work realiably in udma),
and the fact that both the ide hardware and ide disks appear to be
significantly less reliable than it's scsi counterparts (I have had three
maxtor disks fail out of this system in 12 months, thank god it's raid5!.
And I had my secondary interface on my motherboard go out early in the
process... it actually kept working but was slowly corrupting all my data.)
So I just wanted to caution anyone thinking of building a very large ide
system that there is a lot of blood, sweat, and tears involved. Not that I
was unsatisfied with the driver situation.
> I have someone doing four (4) Ultra66's plus the PIIX4 for a RAID 5
> thingy.
That's excellent! Is this supported by the unified ide patch by default?
I'd be interested in hearing what one has to do to get > 6 interfaces going.
And, since I have you here... Do you have support planned for the HPT-366
udma-66 chip? I don't know if it's available on anything else, but it's
coming on abit's new BP-6 and BE-6(?). I just got two of the BP-6's... they
come with both a standard Intel PIIX4 chip on board, and this HighPoint
HPT-366 chip. You can use both at once, seems like a deal for us ide folks
;)
Tom
