How Wide Can Oracle ASM Stripes be and other Cache-Centric Storage Array Tales

[Alzhy]

Cache-Centric Arrays like Hitachi Tier-1 arrays and derivatives (aka XP12000, XP24000) always have that golden rule to stripe wide and "thin".

Since the switch to ASM storage - we've seen quite a substantial toll from our Database Servers to our aptly provisioned XP array but at the same time saw generally a substantial increase in performance and scalability.

There are however environments that have grown substantially and since our LDEVs (or LUN sizes) are ~50 Gigs each - we now have ASM Disksets (under VxVM as raw unitary volumes of course per Symantec/Oracle Best Practice) that could exceed 1024 disks.

My question -- how far wide can ASM striped I/O continue to scale? Will there be penalties growing even wider with such smaller disks? Will it intoruduce cache incoherency at some point that we'll have to do "cache partitioning" for each large DB Server?

I've heard rumours of overly striping accross many front-ends and accross many LUNS can sometimes result in flooding the cache on certain cache-centric arrays -- True or Myth?

TIA for your views... perceieved or proven.

[Sameer_Nirmal · Accepted solution]

Oracle 11g ASM Scalability

63 disk groups
10,000 disks
1 million files per disk group
4PB per disk
Max DB size:
with external redundancy 140PB
with normal redundancy 42PB
With high redundancy 15PB

For XP12000 with ASM, you may want to review the white paper and webinar available at
http://h71028.www7.hp.com/enterprise/us/en/solutions/storage-customer-focused-testing-oracle.html

http://www.oracle.com/technology/products/database/asm/pdf/HP-UX%20-ASM-StgWorks-MP%2002-06.pdf