> > > IDE.
> > 
> > how many channels does it offer per card, does it support RAID-5, does it
> > have onboard memory and how much does it cost?
> 
> different models: 2 channels (U$ 150), 4 channels (U$ 260),
> or 8 channels (U$ 430).
fairly expensive but still $430 USD is a price of only one 18GB SCSI
drive. 
> all support raid 0 and raid 1.
> 
> the 4 and 8 drive versions support raid5, raid1+0 (mirrors, then
> stripe the resulting mirrors), and a hot-swap drive.
> 
> raid5 sucked at writing. e.g, 20MB/s writing raw disk, 20-25MB/s
> writing to raid0, raid1, raid1+0, 5MB/s writing to raid5. reading was
> pretty good on all, and scaled up as the number of drives became
> available. i was getting ~ 20MB/s off 1 raw disk, up to 45MB/s off 4
> disks in raid1+0.
looks like it has no/very little onboard memory or stupid firmware if it
can't do linear write to raid5 fast. or maybe it can't checksum fast.
> in my experience of 12 months with s/w scsi raid, 5 months running
> s/w ide raid, then testing h/w scsi raid and h/w ide raid, there
> is a NOTICABLE difference between software IDE raid and the escalade
> hardware IDE raid. Yes, i had each drive on separate ata66/100
> controllers with software raid (with no slaves). the hardware raid
> felt less `sticky'/`sluggish' under transactional load, which is what
> most unix access is (rather than raw throughput).
agree.

> 
> I'm sure Thor Simon has posted on the lists in the recent past
> about why IDE, even ATA100 IDE, sucks with > 1 device.
with >1 device total or >1 device/channel or per controller?