Testing data usage patterns
From: Don Seiler <don_at_seiler.us>
Date: Thu, 5 Dec 2013 15:58:06 -0600
Message-ID: <CAHJZqBC2jm4aBVC+9X8inoepg-ih86GeDyR85X+LLMB9osF_Zw_at_mail.gmail.com>
Good afternoon. Oracle 11.2.0.3. Currently on ASM but looking to migrate to the Oracle ZFSSA. The big selling point of the ZFSSA was the ARC and L2ARC caches backed by DRAM and flash. However we want to know if our application needs will be met here. We suspect that a fair amount of our I/O is for unique lookups that aren't repeated that often, and so likely would not be in the ZFSSA cache on subsequent executions.
Date: Thu, 5 Dec 2013 15:58:06 -0600
Message-ID: <CAHJZqBC2jm4aBVC+9X8inoepg-ih86GeDyR85X+LLMB9osF_Zw_at_mail.gmail.com>
Good afternoon. Oracle 11.2.0.3. Currently on ASM but looking to migrate to the Oracle ZFSSA. The big selling point of the ZFSSA was the ARC and L2ARC caches backed by DRAM and flash. However we want to know if our application needs will be met here. We suspect that a fair amount of our I/O is for unique lookups that aren't repeated that often, and so likely would not be in the ZFSSA cache on subsequent executions.
However I'd like to be able to determine for sure. I'm wondering if any list members have any suggestions on how to determine how frequently data is read and re-read so we can make realistic expectations on what the performance will be like for our application usage.
We're looking at using RAT to capture a typical workload and then play it back on a "sandbox standby" that we have running on the ZFSSA now, but we have to wait for a downtime window to do some re-linking. Wondering if there are any alternatives that we can also look at.
--
Don Seiler
http://www.seiler.us
--
http://www.freelists.org/webpage/oracle-l
Received on Thu Dec 05 2013 - 22:58:06 CET