RE: RAC in NAS
- From: "Kevin Closson" <kevinc@xxxxxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Wed, 26 Jul 2006 09:13:42 -0700
>>>We are running SUSE 8 Linux 2.4 kernel on Intel i 686 boxes
>>>+ Netapp Filer + 9.2.0.6 RAC NFS mounted.
>>>
>>>We think this gives us bad I/O performance with db file
>>>sequential reads as top waiting event, because the same
>>>queries in the database cloned on a non-RAC non-noac mounted
>>>NFS run much faster.
That is because without noac you are cached on the NFS client. Memory
is faster than ethernet. But, you didn't quantify the wait event with
a time measurement.
Be aware that "Intel I 686 **boxes**" (plural) have more I/O bandwidth
than the filer. Crack open the filer and you'll see a little
intel box (depending on model). So all I/O requests have to go
over an northbridge chipset and out over PCI ethernet. A two node RAC
cluster will always be able to swamp a single-headed filer. Single
headed NAS
(filers) are a bottleneck. Measure sequential writes some time and
you'll
see what I mean.
>>>Oracle blames Netapp and Netapp blames Oracle
Of course they do. That is Oracle standard procedure, and NetApp
does it because they know large sequential I/O is the bane of a
single headed filer--so they need an excuse.
More info:
http://h71028.www7.hp.com/enterprise/downloads/15650%20NAS%20Oracle%20WP
%204A2.pdf
--
//www.freelists.org/webpage/oracle-l
Other related posts:
