Volume Manager - Striping -> sequence writing performance

Typical setup of a Oracle database server could be:

Systemdisk mirrored.

10 disks striped and mirrored as RAID 0+1 for the database-volume.

I have used the command: "time mkfile 1000m testfile" to test the sequence write performance. I wonder why I get performance of c.a. 60 sek when I write to the systemdisk (one mirrored disk), but only c.a. 38 sek when I write to the database-volume (5 striped disks mirrored with 5 other) !?

Why don't I get c.a. 4-5 times performance increase (c.a. 13-15 sek) ?

Do you have the same sindrome on your servers ?

Is this normal and then why ?

Regards

Halldor Gretar

p.s.

This sindrome is on all servers that I have tested, both setup by me and others, and both with Veritas Volume Manager and Sun DiskSuite.

[820 byte] By [hge1] at [2007-11-25 23:11:54]
# 1

Hello,

I齧 not an expert on volume management, but maybe these considerations that come to my mind can help you to improve your performance:

1.- The interlace size of the striping. You should adjust the size of the striping to match the I/O requests made by the Operating System or by the database management software (is the data access in a raw mode?). For example, if the data access is made through normal ufs access, the stripping size should match the block size of the file system.

2.- Are those disks on different controlers? Maybe a saturation of the controler, of the bus, etc... could slow down your I/O read/writes.

Bye,

jmiturbe

jmiturbe at 2007-7-5 18:01:51 > top of Java-index,Storage Forums,Storage General Discussion...