[lug] software RAID
Dean Brissinger
brissing at kaidok.com
Sat Feb 11 14:56:44 MST 2006
On Friday, February 10, 2006 10:20 PM D. Stimits wrote:
> Jeff Schroeder wrote:
>> Hugh wrote:
>>> read something that said putting them on the same ide interface is
>>> worse than bad. If so, is it only performance related or do
>>> terrible things happen (guess who put both drives on ide1).
>>
>> 1) Performance, as you said. Since you're writing to both drives in
>> a RAID-1 configuration, you'll be crowding the bandwidth on the IDE
>> channel. Using two channels will allow you to have greater speed.
>>
> Worse than that: IDE does not allow two drives simultaneous access.
> Maybe SATA is better, but I don't know. In terms of SCSI, there is a
> fast burst then the drive disconnects from the bus. SCSI has no
> problem dealing with maybe 7 drives on one cable (except perhaps for
It's very bad:
In the case of a mirror putting two devices that would be activated at
the same time will dead-lock many systems. If not dead-lock it will run
very very slow. The reason being IDE controllers do not provide a cache
for events that would allow both devices to alternate control of the bus
(like SCSI does). Take the case you put a Zip drive and a system drive
on the primary IDE bus as master and slave. Everything will be fine
until you copy a file from the system drive to the zip drive. At which
point the system will a) need swap, b) need dedicated i/o to the hdd for
read, and c) need dedicated i/o to the zip. The controller will refuse
two and the end result is a deadlock after one or two (lucky) cached
writes. The solution is to add IDE busses. Check out some IDE/PCI
adapters. Many conflict w/ the system bios in old hardware though.
SATA does not address this. SATA raids are built with one SATA bus per
disk and combined with a fibre channel or SCSI backplane. You can
actually get chassis now that can interchange SATA and SCSI disks in the
same slot because of this design.
Also, modern SCSI can talk to many more drives than 7 on a single chain.
All SCSI/FibreChannel hard drives have a tuneable option for how long
they hold the bus after they are done talking. In build-your-own
environments it is required to tune this to zero to prevent bus
overruns/cache overflows. Though, I don't recommend building such a
thing home-grown. If you need something like that go buy a JBOD chassis
with a solid silicon backplane for it. ;-)
More information about the LUG
mailing list