Matt Mc_Carthy -> Leonard Erickson >>IDE twins.<< {at} 05 Jun
03  23:21:58

 LE>> If they are on different channels, then you can read from one and
 LE>> write to the other at the same time.

 MM> Not quite Leonard.  Regardless of multitasking, or whatever people want
 MM> to call it from day to day, the CPU can only run ONE task at a time,
 MM> period!

  That's right. But you can have more CPUs in a computer. :) And as far as
I know, my 750 MHz AMD Duron can run 3 integer instructions simultaneously
in the best case. But that's not the point.
  I don't think the CPU is the bottleneck since bus master DMA ATA became common.

 MM> You will NEVER see both HD lights on at the same time.

  Yeah, I have only one HDD LED on the case I can see. :)

[copying from one disk to another]

 MM> In the case of BOTH drives on the same channel (cable), The CPU has to
 MM> first select the channel, then read to main RAM, then write that RAM to
 MM> the second drive on that same channel.

 MM> In the case of the drives on separate cables (channels), the CPU has to
 MM> select the _read_ drive, read the data into main RAM, then select the
 MM> _write_ channel, then write that data to the second drive, and repeat
 MM> until the copy operation is completed.

 MM> By having both drives on the same channel, the CPU has to perform two less
 MM> _select_ options per data block transfer, as that one channel simply
 MM> remains selected.

  Hmm. Would you elaborate what exactly you mean by selecting a channel?
(You can even post assembly.:)) I know that 2 controllers have 2 port
ranges, but I don't see how that slows down anything. On the other hand,
they can use different interrupts, which can help performance slightly.

 MM> All in all, HOW you use the computer will determine what gets connected
 MM> where for optimum results.

  OK. Since I have only one ATA cable that can do > UDMA mode2, I use
that with both HDDs.

  Here's some profiling in GNU/Linux... It's a real life test on an idle
system, so not exact. :) It's with default bdflush settings, so there's
about a 2 seconds write delay.

VP_IDE: VIA vt82c686b (rev 40) IDE UDMA100 controller on pci00:07.1
    ide0: BM-DMA at 0xd000-0xd007, BIOS settings: hda:DMA, hdb:DMA
    ide1: BM-DMA at 0xd008-0xd00f, BIOS settings: hdc:DMA, hdd:pio

1. drives on the same cable

hda (ide0/master) is SAMSUNG SV1824D that can do UDMA4
hdb (ide0/slave) is Maxtor 6Y080L0 that could do UDMA6 (well, not with this
controller...) set to UDMA2 for the sake of comparison

"hdparm -t" results:
/dev/hda:
 Timing buffered disk reads:  64 MB in  2.85 seconds = 22.46 MB/sec
/dev/hdb:
 Timing buffered disk reads:  64 MB in  2.16 seconds = 29.63 MB/sec (This
would be 47.41 MB/sec with UDMA4.)

Copying 204800000 bytes from hda to hdb 3 times: (sync; time { dd
if=/dev/hda of=/mnt/hdb/tmp/prof/0.dmp bs=8192 count=25000; sync; })

real    0m14.026s
user    0m0.020s
sys     0m1.270s

real    0m13.908s
user    0m0.030s
sys     0m1.280s

real    0m14.134s
user    0m0.000s
sys     0m1.340s
 
"real" is the elapsed time
sys+user is CPU-time

2. drives on separate cables

hda (ide0/master) is SAMSUNG SV1824D
hdd (ide1/slave) is Maxtor 6Y080L0 (it should be master, but who cares if
it works?) connected with an older 40-wire cable
"hdparm -t" results:
/dev/hda:
 Timing buffered disk reads:  64 MB in  2.85 seconds = 22.46 MB/sec
/dev/hdd:
 Timing buffered disk reads:  64 MB in  2.34 seconds = 27.35 MB/sec
  (This is a bit slower than was on the other cable. I guess it's because
of UDMA transmit errors. Any other idea?)

Copying 204800000 bytes from hda to hdd 3 times:

real    0m11.052s
user    0m0.010s
sys     0m2.200s

real    0m11.016s
user    0m0.020s
sys     0m2.240s

real    0m11.000s
user    0m0.020s
sys     0m2.060s

  So when the drives are on separate cables, the copying is faster, even
with that handicapped disk read statistics. But the CPU utilization
increased as you stated, so I look forward to your elaboration. :)

  PS. Good thing that I opened the case. The cooler on the video card
wasn't working. :(:
  I wish I could switch off the power source of the devices individually.
Even the DDS drive I only use once a month is hot. :( At least the Maxtor
drive goes to standby when not used.

--- SkyReader/Linux v1.01a-7