[Techtalk] Software vs. Hardware RAID

[Sophie]

On Tue, Aug/20/02 08:38:37AM -0700, Dave North wrote:
> 
> Bear in mind that most references I've seen imply that putting two disks
> on one IDE channel will bog things down enough to more than overcome any
> advantage of striping (I haven't done that test, I regret to say, so I
> don't know if it's true from personal experience).

That depends on your controller. With modern controllers (and Andrew's raid card certinaly falls under this category, as it's ata/133), the cables are 80-wire (as opposed to the older 40-wire cables). The extra wires are used for ground lines, in a similar was to lvd scsi (if I understand the technologies correctly), which means there is less unknown error, therfore less allowances need to be made between data, therfore everything can move faster.

>From what I understand, this means that two ide disks on an ata/100 (or ata/133) bus will not trip up each other as often as two drives on ata/66 (or ata/33 (!)), allowing for speed differences. I could, of course, be totally wrong :)

> 	One thing I've never seen is a discussion of whether it matters if
> one of those disks is unused: for example, hda1 is a hard drive and hda2
> is one of the CDs, usually not even mounted. I would assume that safe, but
> do not know for sure.

This is certinaly true with older (ata/66/33) systems - you'd have one drive sharing with a cdrom, and the other drive sharing with the other cdrom, otherwise the two drives would be transfering too much data for the ide bus, which would cause a bottleneck and slow things down.


> > 	Would setting the jumpers on the 4 drives to slave and master
> > instead of cable select, go toward solving my problem?
> 
> I have never used or experimented with cable select, but have read that it
> default cuts off dma transfers. Have you seen the program hdparm? If so,
> once a cable-select disk is mounted, is it using dma?

Intresting. I'll try that when I get chance. hdparm is a wonderful thing :)

- sophie