RE: RAID10 and table layout

I agree with Art on striping. If you can, make the mirror pairs between different disk controllers (assuming you have at least 2 controllers). This is good for redundancy and controller failure. (Had a Sequent once with 1 controller which failed and wiped all disks - even though they had mirrors). I have found stripes over 4 disks to be effective. (One site with plenty of spindles I used stripes of 6 disks). Analyse your database(s) by groups, with rootdbs being a group and llogdbs being another and your database may be 1 or 2. Plan your groups and stripe your disks accordingly. I have found that rootdbs will often need to share the same disks as a chunk for the database. The database groups may be high use tables versus low use tables or it may be OLTP used tables versus (low use large volume) history tables. I have planned to create the active tables first. The first 2 chunks become the most active. Hence create these chunks as stipes along the middle of the disks. Later chunks with less used tables (well initially anyway), use the outside parts of the disk. HTH Murray Wood -----Original Message----- From: Art S. Kagel [SMTP:kagel@bloomberg.net] Sent: Tuesday, July 20, 1999 9:53 AM To: informix-list@iiug.org Subject: Re: RAID10 and table layout The best way to do this is to mirror the drives themselves one-to-one and in that the mirrored pairs have to be the same drive. I'd suggest getting 12 4GB drives to create 6 mirrored pairs of matched drives (different manufacturing lots on each side of the pair for safety). Then you can tape 2 pairs for rootdbs, logs, and low activity tables and stripe the remaining 4 pairs for performance giving 16GB of striped space. Just one layout of course. Remember "Wider is better" a wider stripe (more spindles) will outperform a narrower one made up larger faster drives. Art S. Kagel "Carlson@WHSmith" wrote: > > IDS: 7.30.uc7 > OS: HPUX 10.20 > > I've just received word that I'll be able to implement striping / > mirroring for my Informix system; total database size is about 25G. I'm > looking at 3 9G drives and ?? 4G drives on a mirror. Any ideas, > comments, generalizations? I'm a bit new at this, as we've always used > singelton drives ( 2G / 4G ). Any recommended reading? > > Thanks in advance > John Carlson > Informix DBA > WHSmith USA

Murray Wood wrote: > I agree with Art on striping. > > If you can, make the mirror pairs between different disk controllers > (assuming you have at least 2 controllers). This is good for redundancy > and controller failure. (Had a Sequent once with 1 controller which failed > and wiped all disks - even though they had mirrors). > Also good for performance - if the mirror is on the same bus it uses twice as much of the bus' bandwidth for every mirrored write. > Snip > > I have planned to create the active tables first. The first 2 chunks > become the most active. Hence create these chunks as stipes along the > middle of the disks. Later chunks with less used tables (well initially > anyway), use the outside parts of the disk. By middle do you mean the middle of the cylinder range with the high and low cylinder numbers being considered outer? The theory of that is that to get the head to the middle of the cylinder range is at most only half the width of the disk surface from where the head is now but to get to the centre of the disk the head might have to traverse the entire width. I understand that Pick put the directories in the middle of the cylinder range for the same reason. Regards Ian