File Disk Layout Opinions/Comments......

Hello, IFX 10 UC4 Redhat AS3 HP Hardware 2 controllers I have 256 GB of data stored on a system configured to have: 6 mirrored disks (RAID 1) with each disk containing table data, indexes, TX logs and temp space. I am contemplating setting up 3 disks to contain database tables and 3 disks containing tx logs and temp space (or some variation thereof...). My reasoning is as follows: With the current disk setup there will be a lot of thrashing going on as my disk heads are flying all over the disk trying to write to txlogs, write to temp space and read from my tables. My goal is splitting the functionality so one set of disks will be doing primarily writes and one will be doing primarily reads. Is this a valid plan of action? If not why? Suggestions for further improvement? Thanks, Tam.

Tam O'Shanter wrote: > Hello, > > > IFX 10 UC4 > Redhat AS3 > HP Hardware > 2 controllers > > I have 256 GB of data stored on a system configured to have: > > 6 mirrored disks (RAID 1) with each disk containing table data, indexes, TX > logs and temp space. > > I am contemplating setting up 3 disks to contain database tables and 3 disks > containing tx logs and temp space (or some variation thereof...). > > My reasoning is as follows: > > With the current disk setup there will be a lot of thrashing going on as my > disk heads are flying all over the disk trying to write to txlogs, write to > temp space and read from my tables. > > My goal is splitting the functionality so one set of disks will be doing > primarily writes and one will be doing primarily reads. > > Is this a valid plan of action? > > If not why? Sounds good to me! > Suggestions for further improvement? I'll go with my favorite. Consider striping each of the three pairs into two three pair RAID10 arrays with a small (~16 or 32K) stripe block size. You'll always be getting 3-6 spindles spinning for every read/write and total throughput should improve unless some of the data on one pair is far more intensively accessed than the data on the others. Also consider keeping the controllers for the two arrays independent to improve throughput even more. Art S. Kagel > Thanks, > > Tam. > >