On Sun, Aug 29, 1999 at 12:37:45AM -0700, Skip Harrison wrote:
> Thanks for the info. But are you using a Promise Ultra66 or and Ultra33?
Ultra66
> Does the Promise card pre-boot screen show the card and drives are in UDMA
> mode 4?
Yes
> Mine does, but I still do not think it is carrying over to Linux
> since if I shut off UDMA66 with the IBM program, put a regular 40 conductor
> cable on the card/drive, or connect it to the regular IDE interface on the
> motherboard (UDMA33) I still get 17MB/sec. I would think I should get
> something like you mention below, 23-24MB/sec, double the plain jane WD 4gig
> DMA33 drive.
I believe I'm actually getting UDMA66 speeds in linux primarily because
the driver says so:
hde: ST317242A, 16446MB w/512kB Cache, CHS=33416/16/63, UDMA(66)
(that was for a 5400rpm drive; the only example I could grab right now,
but the ST328040A also produced a similar message).
Given that none of the current drives can sustain rates much greater
than 20MB/s (and most probably hit in the 10-15MB/s range), I don't think
it's surprising that there's no big difference in performance between
burst speeds of 33MB/s and 66MB/s.
So why has the world been blessed with UDMA-66? I think, aside from
marketing, which has been rather subdued in this case, it's a matter
of looking forward. The density and rotational speed of a platter is
increasing very predictably, and this is what drives the sustainable
transfer rate. You can see it won't be long before even cheap drives
will exceed the UDMA-33 speeds. Moreover, there are other overheads
involved in I/O besides the burst DMA transfer rate. The total time
to read a series of sequential sectors includes also the time to
handle the interrupts and initiate each new request. So if you are to
match the drive's maximum sustainable rate, the DMA rate must be faster.
> >I didn't do any of that. I do have CONFIG_IDEDMA_AUTO=y and
> >IDEDMA_PCI_EXPERIMENTAL=y, CONFIG_BLK_DEV_IDEDMA=y,
> CONFIG_BLK_DEV_PDC202XX=y,
> >and PDC202XX_FORCE_BURST_BIT=y.
>
>
> I thought the PDC202XX_FORCE_BURST_BIT=y was only for the Ultra 33 card. At
> least that was what I gathered from reading the help during "make
> menuconfig" and in the PDC202XX.c file so I did not turn it on. Maybe I'll
> give that a try. I also have the other options on, but not the
> FORCE_BURST_BIT.
PDC202XX_FORCE_BURST_BIT is only important if you have more than 2
promise cards. It works around a Promise BIOS bug that's present in
both the ultra33 and the ultra66.
> Where does one get the bonnie program? I have seen it referenced, but could
> not track it down...
There is a web page. I forget where it is. It wasn't hard to find
with a web search.
Jan Edler
NEC Research Institute
