[Pc_Support] External SATA bracket ($9), external SATA enclosure/power ($35)

[Bryan J. Smith]

Jason Boxman <jasonb at edseek.com> wrote:
> That's not all that was said.  It was also discussed that
> having two disks on the same ATA cable and moving data between
> each disk doesn't suffer much of a penalty versus the source
> and target disks being on different ATA ports altogether.

Huh?  It very much does.

About the only thing I can think of why it wouldn't is if the
combined rate of write (from one disk) and read (to the other) is
greater than the PCI bandwidth available.  This _could_ be very much
the case with drives today on 32-bit @ 33MHz PCI busses (133MBps
theoretical).

In that case, the bottleneck is now the PCI bus, and since _both_
configurations have to traverse it (to memory), the separate channels
make _far_less_ impact.  Of course, if you've got PCI-X or PCIe
things change once again.  ;->

> Yes, I understand that.  So what's the practical overhead of
> having two ATA devices in UDMA mode 5 on the same port copying
> data amongst each other?  

The fact that the system only sees *1* hard drive at a time is the
problem.  That means one drive transfers to memory, the ATA channel
is reconfigured, and then it transfers back to the other disk from
memory.

With separate ATA channels, the system _always_ can read/write to
_either_ drive -- directly.  No ATA channel changes.  That overhead
can be signficant and take time away from transfer.

> What about if you had two mainboard ATA ports, with four disks,
> and you had two RAID 1 mirrors and you RAID 0 stripped them?

Bend over.  ;->

There is this farce that ATA devices on the same cable can talk to
each other.  ATA is not SCSI.


-- 
Bryan J. Smith     Professional, Technical Annoyance
b.j.smith at ieee.org      http://thebs413.blogspot.com
----------------------------------------------------
*** Speed doesn't kill, difference in speed does ***