ASM stripe width in 10g

Will the real stripe width in ASM please step forward?

I've been reading conflicting views on the stripe width for ASM recently. Someone please set me straight on this.

First, I heard that ASM performs its striping at the extent level. In this case, all the extents would be spread out over all the disks in the disk group.

I can create a tablespace with a uniform extent size of whatever I want. Does this mean that I'm changing the stripe width as well? 10g also allows you to create a dictionary managed with ASM. This seems like a very bad idea because if you set a pct increase of 50, you probably won't have very many extents, and your data will not be very balanced.

I've also read that ASM stripes the data according to the template setup for the specific data file (v$asm_template). Inside the template, you can specify only one of two stripe sizes, coarse or fine. Coarse being 1MB, fine being 128k. It seems that this would be the way to go. In this case your stripes would be one of these sizes regardless of your extent size.

Have the documents I've read just been blowing smoke about how ASM stipes the data at the extent level? Maybe I'm missing something.

Just realized, perhaps they are just saying that each extent is striped at the stripe size indicated in the template. Anyway, it's not very clear.

Additionally, ASM claims to prevent fragmentation. I don't buy this either. If fragmentation refers to the sizeable space unavailable in a datafile because no objects can allocate their next extent inside it, and I can still have a dictionary managed tablespace using ASM, I could still find a way to get some very fragmented data files.

Thanks in advance,
Kevin

To reply by email, remove "_rm" from email address. The address was altered to prevent spam. Received on Wed Aug 04 2004 - 15:11:28 CDT