We're building a new non-RAC 188.8.131.52 system on x86-64 RHEL 5.7 with ASM diskgroups stored on a NetApp device (don't know the model # since we are not storage admins but can get it if that would be helpful). The system is not a data warehouse--more of a hybrid than pure OLTP or OLAP.
In Oracle® Database Storage Administrator's Guide 11g Release 2 (11.2) E10500-02, Oracle recommends using allocation unit (AU) size of 4MB (vs. a default of 1MB) for a disk group be set to 4 MB to enhance performance. However, to take advantage of the au_size benefits, it also says the operating system (OS) I/O size should be set "to the largest possible size."
Since we're using NetApp as the underlying storage, what should we ask our storage and sysadmins (we don't manage the physical storage or the OS) to do:
* What do they need to confirm and/or set regarding I/O on the Linux side
* What do they need to confirm and/or set regarding I/O on the NetApp side?
On some other 184.108.40.206 systems that use ASM diskgroups, I checked v$asm_diskgroup and see we're currently using a 1MB Allocation Unit Size. The diskgroups are on an HP EVA SAN. I don't recall, when creating the diskgroups via asmca, if we were even given an option to change the AU size. We're inclined to go with Oracle's recommendation of 4MB. But we're concerned there may be a mismatch on the OS side (either Redhat or the NetApp device's OS). Would rather "first do no harm" and stick with the default of 1MB before going with 4MB and not knowing the consequences. Also, when we create diskgroups we set Redundancy to External--because we'd like the NetApp device to handle this. Don't know if that matters regarding AU Size.
Hope this makes sense. Please let me know if there is any other info I can provide.
Unless there is something else, unusual, about your system my recommendation would be that you stay with the default 1MB.
This is not dissimilar from the theoretical gains to be had from changing the default block size from 8K. Perhaps in some rare system it helps but usually it just introduces the DBA team to creating more SRs due to finding bugs.
Thanks Dan. I suspected as much due to the absence of info out there on this particular topic. I hear you on the comparsion with deviating from a tried-and-true standard 8K Oracle block size. Probably not worth the hassle. I don't know of any particular justification with this system to bump up the AU size--especially if this is an esoteric and little-used technique. The only justification is official Oracle documentation suggesting the value change. Since it seems you can't change an ASM Diskgroup's AU size once you create it, and since we won't have time to benchmark using different AU sizes, I would prefer to err on the side of caution--e.g. first do no harm.
Does anyone out there use something larger than a 1MB AU size? If so, why? And did you benchmark between the standard size and the size you chose? What performance results did you observe?