Using FusionIO for Oracle datafile storage

From: Paresh Yadav <yparesh_at_gmail.com>
Date: Fri, 13 Jul 2012 00:40:56 -0400
Message-ID: <CAPXEL0JZeF7DyiifUUiAdHSCp9+qtJZifre81Rz6pY+o42Mguw_at_mail.gmail.com>

I am planning to use fusionIO for a database that requires high random read IOPS (100,000+ IOPS). I had some questions if you can provide some guidance and also point me to some good material.

• If we decide to put our data (tablespaces/data files) on FusionIO do we need to replicate it to a regular hard drive based storage? I know fusionIO is nonvolatile memory but is it reliable (enough) to store data on it for the long term? Do we need to replicate? If yes what did you use? Will ASM with Normal redundancy diskgroups (one failuregroup on fusionIO and other on regular hard drives) a possible option?
• If we decide to use fusion IO to store only "hot" data i.e. last 90 days of data (500 GB+) and rotate the data by moving "oldest day" out and "latest day" in logic, we will incur the read-erase-write overhead with fusionIO. We can live with that. My question is, is there any limitation on how many such "read-erase-write" cycles fusionIO memory can handle before the performance (IOPS) degrades and/or the memory becomes unreliable? I am under the impression that this will happen with any SSD storage but I might be mistaken.
• With fusionIO, do we need to reserve part of the storage capacity to work around read-erase-write overhead or fusionIO does it internally and we get to use the entire advertised storage capacity?
• Am I on the right track!? ;)
• Any other thing that I should watch out for?

-- 
Thanks
Paresh
416-688-1003


--
http://www.freelists.org/webpage/oracle-l

Received on Thu Jul 12 2012 - 23:40:56 CDT