Now that a write to Flash cache is a commit to the database I assume that Flash cache must be fault tolerant to a single FMOD failure.
Does this mean that now mirror copy writes which are sent to a separate failure group must also be written to flash cache ?
Secondly in the original Flash cache whitepaper of 2009 http://www.oracle.com/technetwork/middleware/bi-foundation/exadata-smart-flash-cache-twp-v5-1-128560.pdf
"While the default behavior for sequential scans is to bypass the flash cache, this is not the case when KEEP is specified. If KEEP has been specified for an object, and it is accessed via an offloaded Smart Scan, the object is kept in and scanned from cache. Another advantage of the flash cache is that when an object that is kept in the cache is scanned, the Exadata software will simultaneously read the data from both flash and disk to get a higher aggregate scan rate than is possible from either source independently."
Is this still valid in the world of writeback flash cache or can we assume that this is no longer a necessity. I will be doing some experimentation on this but if anyone already knows the answer please do tell.
Yes to both questions.
1. Write to flash, when i writeback mode is mirrored.
2. The rules governing what is kept in flash are unchanged. Therefore, large IO's are not kept in flash, unless marked as KEEP.