metadevice striping

[thomgreen]

I have a production system on which I must load a new OS. I will move my prod. app. and db to my development box. On the dev box I am mirroring but not striping across disks & controllers. Our app. sort of a hybrid OLAP/DSS tool.

In general should I expect to see performance deteriorate?

 

[jimbopalmer]

The only point to striping is for disk access performance, so one would expect that the disks throughput would be slower without it. This would affect the app in general to the extent it was disk throughput bound.

If the app is more focused on latency or not disk bound at all, then the effect will be less. I tried to remain child-like, all I acheived was childish.

 

[Guest_imported]

Not enough information.

Building a filesystem structure for a database is 50% science and 50% black art.

If you have a large database with transactions, logging, indexing as well as your raw data and swap space, you will probably do better leaving your disks as single spidles (or mirrors of single spindles) and so your log/transaction/swap/index/data writes are effectively independent. You may find that striping all your disks together then carving them up with a logical volume manager will have a serious performance hit due to the seek times of the disk heads. This is especially true if you are dealing with a large number of small writes.

If you have a simple database structure that does large, sequencial writes, a mirrored stripe (or preferable a striped mirror) may be your best best.

 

[Guest_imported]

Not enough information.

Building a filesystem structure for a database is 50% science and 50% black art.

If you have a large database with transactions, logging, indexing as well as your raw data and swap space, you will probably do better leaving your disks as single spidles (or mirrors of single spindles) and so your log/transaction/swap/index/data writes are effectively independent. You may find that striping all your disks together then carving them up with a logical volume manager will have a serious performance hit due to the seek times of the disk heads. This is especially true if you are dealing with a large number of small writes.

If you have a simple database structure that does large, sequencial writes, a mirrored stripe (or preferable a striped mirror) may be your best best.

 

[Guest_imported]

Not enough information.

Building a filesystem structure for a database is 50% science and 50% black art.

If you have a large database with transactions, logging, indexing as well as your raw data and swap space, you will probably do better leaving your disks as single spidles (or mirrors of single spindles) and so your log/transaction/swap/index/data writes are effectively independent. You may find that striping all your disks together then carving them up with a logical volume manager will have a serious performance hit due to the seek times of the disk heads. This is especially true if you are dealing with a large number of small writes.

If you have a simple database structure that does large, sequencial writes, a mirrored stripe (or preferable a striped mirror) may be your best best.