On Tue, 6 Apr 1999, Stein Gjoen wrote:

        Hi,

> In the Multi Disk HOWTO there are already some notes on
> optimum use of disks and RAID levels that you might find
> of interest.
>       http://www.nyx.net/~sgjoen/disk.html
> 
> All feedback is welcome; I wrote it.

        I have a little suggestion regarding the use of raid that involves
IDE. I know you already know that, but i think it can be said clearlier:

        1.- If you do IDE raid, use both ports (eg, raid0 or raid1 on
hda1 and hdc1). Reason follows.
        
        This is obvious. What maybe isn't obvious, because is somewhat a
odd setup, is a 0+1 IDE raid. Surelly it won't win any raid award of the
year, but maybe somebody will aim those targets:

        1 - minimum cost
        2 - maximum storage
        3 - redundancy
        4 - the best possible performance

        It's a high -readily available (no tapes) cheap mass storage
solution that suits some requirements, e.g. a Intranet repository.

        Also, an SMP machine will behave (as mine) surprisingly well.

        Having a 0+1 IDE raid of 20Gb cost less than $530. I'd use scsi 
for the system disc, however. But SCSI II is also cheap nowadays. 

        One could wonder (as I posted sometime ago) how to setup the 0
raids. Do i make raids 0 on the master/slave device of the same IDE port,
and mirror both "ports"  later? or do i do raids 0 on master/master
slave/slave devices and later mirror them? 

        The answer is keep raids 0 on the same controller, i.e. raid0 A
hdaX-hdbX, raid0 B hdcX-hddX and later mirror them. 

The reason is that on a master-slave setup, the slave disc is controlled
by the master. If on a [hda-hdc]+[hdb-hdd] raid 0+1 the master device of
*any* IDE controller fails, the slave will inmediatly fail also (I'd bet
it surelly will happen if it's the slave who fails, so this statement
could be widened to "if *any* disc fails"), rendering your raid 0+1
inmediatly unusable, and making recovery thougher.

        Just my 2 cents.

*****---(*)---**********************************************---------->
Francisco J. Montilla              Systems & Network administrator
[EMAIL PROTECTED]      irc: pukka        Seville            Spain   
INSFLUG (LiNUX) Coordinator. www.insflug.org   -   ftp.insflug.org

Reply via email to