Re: [TCLUG:12134] SCSI vs. IDE

[<Nick.T.Reinking@supervalu.com>]

>Now these ata66 controllers, how do they make the landscape different?  It
>seems like they have scsi-like drivers in windows.. dont know about linux
>support.

Well, ATA/66 is basically ATA/33 but with a maximum transfer rate of 66MB/sec, 
and higher data integrity.
This "data integrity" depends on having a special (i.e. expensive!) 
80-conductor IDE cable.   Instead of having
a normal 40-pin IDE cable, you have to use an 80-pin IDE cable.  It's basically 
the normal IDE cable except
for every other wire on the cable being a ground, and having one wire punctured 
at a special (28th wire?)
spot.  It's very unusual - and the cables are *very* expensive - even more so 
that UW internal SCSI cables.
Now you IDE  folks now what it's like for us who use SCSI.  ;)

> Your comment about CPU indicates that SCSI is better for even
>one drive.  This mirrors an experience I had when I was running a heavy
>test that read and wrote random 8 meg segments of a 64 meg file.  My
>machine slowed down to a crawl and all the CPU time was being spent on
>'system'.

Well, IDE can be very CPU efficient - *if* DMA is enabled for your drives.  (It 
has to be compiled into the
kernel).  If your IDE controller isn't using DMA, you're limited to PIO 4 
transfer rates - which can be limited
either by the maximum transfer rate (modern IDE drives can hit that limit 
pretty easily), or by the amount
of CPU power you have - pushing about 17MB/sec can really tax your CPU, so it 
might have been a CPU
barrier that you ran into.

Be sure to compile in DMA - or get SCSI!

Nick Reinking