In a
typical RAID 10 network, you have two servers, primary and secondary, both
running RAID 5 with three hard drives apiece.
The first hard drive contains the working copy of the database, the
second is a mirror image of the first, and the third is a parity check for the
first two drives. In this way, if an
error occurs on either or both of the running drives 1 and 2, the database may
be repaired from the parity info on drive 3.
In
production, RAID 10 is used for high-traffic database access, so speed is a
key. The backbone network is usually
fiber optics or CAT6 to the application servers.
A SAN
network is basically designed for the backup-paranoid. Usually a direct fiber optic line from the
center node or “home base” of the production network is linked to a backup
server (or sub-network of backup servers).
There is usually a “flash” backup server to do the data fetching, but
this is usually considered the most “volatile” backup. A tape backup system, a second array of
disks, or several other servers will exist to “backup your backup”.
At the
scheduled time of backup on our RAID 10 network, all disk writes are
temporarily paused, and all transactions are held in memory until the backup
process is over. The “flash” backup
server will then make a rapid disk transfer from the primary database
server. Once this is done, the RAID 10
servers immediately write to disk the transactions saved in memory. The flash server disks are then copied again
to long-term storage, and are prepared to “flash” the next server in the backup
task list.
So, using
RAID 10 with a SAN will have a primary server that mirrors and parity-checks
the database. You also have a secondary
server that duplicates this task, in case the primary server fails, and/or to
balance the transaction load. A backup
sub-network on a timed schedule will flash-mirror the primary server, and once
this task is complete will create another copy of the backup to archive.
Jake Bishop
Midwest
Information Systems, Inc.