This blows because I am porting a legacy application
from an MVS system. This application accesses two
sequential datasets - flat files that are over 10GBs.
Since both files have to be accessed at the same time,
I was hoping to put them on different platters to
avoid disk contention - two processes attempting to
read from the same or near cyclinder groups but since
like you say all heads are linked together, this wont
do much good.
Yeah, pretty much your best bet would be to put each file on its own spindle. Using a RAID controller with a lot of cache may solve the problem for you too, depending on the DB usage pattern.
I know when setting up a large MS SQL server, best practice is to put each log file on its own set of spindles, usually mirrored. Since SQL logs just append to the end of the file, essentially the heads just sequentially move to the next track.
If your project is big enough, you could use SAN or NAS disk, 64gb of disk cache should give you pretty good performance for two 10gb files.
